pySQLbridge

Answer SQL Server's wire protocol convincingly enough that Excel and Power BI connect to a JSON file, a CSV, a workbook, an Access database or an HTTP API and see a database. Read-only, so only the SELECT surface has to hold up.

Status

A real SQL Server client lists the tables, then selects from a CSV file, a JSON file, an Excel workbook, an Access database or a live HTTP API over the wire, with column types, NULLs, WHERE and TOP. Parameterised queries work, which matters because clients send those as RPC calls to sp_executesql rather than as SQL batches.

The SQL covers what a client and a person actually send: joins, GROUP BY with HAVING, DISTINCT, OFFSET/FETCH, CTEs, subqueries and derived tables, CASE, CAST, expressions and aliases in the select list, scalar subqueries, UNION, EXCEPT, INTERSECT, correlated subqueries, window functions, and 96 scalar functions over 10 aggregates. All 1,243 queries in scripts/differential.py answer identically to SQL Server 2025, declare the same kind of column for each answer under the same heading, and where both refuse, refuse with the same message number. Its 175 whole batches agree statement by statement, and a batch that will not compile is refused before any of it runs, with the number a real server gives it. Anything it cannot answer is refused by name rather than answered wrongly.

$ python -m pysqlbridge.server --config examples/tables.json
serving 3 table(s): cities, people, pokemon
warmed 3 source(s) in 412 ms
listening on 127.0.0.1:1337
connection from 127.0.0.1:52434
127.0.0.1:52434 logged in as DOMAIN\user (app '.Net SqlClient Data Provider', database 'master')
127.0.0.1:52434 query: SELECT id, name, score, retired FROM people

> SELECT COUNT(*) AS n, MIN(score) AS lo, MAX(score) AS hi FROM people
n           lo                       hi
----------- ------------------------ ------------------------
          4                     78.0                     99.5

> SELECT TOP 3 name FROM pokemon
name
----------
bulbasaur
ivysaur
venusaur

> SELECT name, score FROM people
name            score
--------------- ------------------------
Ada Lovelace                        99.5
Grace Hopper                       87.25
Edsger Dijkstra                     78.0
Barbara Liskov                     93.75
(4 rows affected)

scripts/run_dev.ps1 serves the example tables and prints the connection strings for sqlcmd, Excel and Power BI.

Installing

pip install pysqlbridge

That puts a pysqlbridge command on the path, which takes the same arguments the module does:

pysqlbridge --config examples/tables.json

Python 3.10 or newer. cryptography comes with it, for the self-signed certificate the login tunnel needs; pyopenvba comes with it, for reading Access databases, and is pure Python with no dependencies of its own; pywin32 comes with it on Windows, for Windows Authentication. Only the last is platform-bound. Every bridge needs cryptography, pyopenvba only reads Access databases, and pywin32 only does Windows Authentication, so a bridge serving files over SQL authentication runs anywhere Python and cryptography do.

The executable below needs no Python at all on the machine it runs on, which is the reason it exists.

Building the executable

.\scripts\build_exe.ps1

Runs the suite, builds dist\pysqlbridge.exe with PyInstaller, then stages the result in a directory with no source tree and drives a real client through it: log in, run a query, list the catalog. About 13 MB, no Python needed on the target, and it takes the same arguments the module does.

It needs the package installed with its build tools, pip install -e ".[dev]", since PyInstaller comes with the dev extra and the spec reads the installed package's metadata. -SkipTests skips the suite, and -Port moves the smoke test off 1399. The smoke test logs in with Windows Authentication, so the build runs on Windows only.

The smoke test is not politeness. PyInstaller cannot see an import that happens inside a function, so a build can start, listen and load its tables and still fail the instant a client authenticates. That is exactly what happened: sspi is imported inside a function in auth.py, and the first build died on win32timezone, which sspi reaches at runtime. Only running the executable found it. Those names are listed in pysqlbridge.spec with a note saying why.

Configuration

The shortest useful config is a base URL:

{
  "discover": [
    { "url": "https://pokeapi.co/api/v2/" }
  ]
}

That crawls the surface and serves what it finds. Against the PokeAPI it produces 27 tables; against dummyjson.com, 8. Naming tables one at a time is still there for the cases discovery cannot reach:

{
  "tables": [
    { "name": "people", "csv":  "data/people.csv" },
    { "name": "sales",  "csv":  "data/sales.csv", "delimiter": ";" },
    { "name": "cities", "json": "data/cities.json" },
    { "excel": "data/budget.xlsx" },
    { "access": "data/club.accdb" },
    {
      "name": "pokemon",
      "http": {
        "url": "https://pokeapi.co/api/v2/pokemon?limit=25",
        "path": "results",
        "ttl": 300,
        "timeout": 20
      }
    }
  ]
}

Both keys may appear. A named table wins over a discovered one of the same name, because a person who wrote a name meant it, and the log says which discovered table it replaced. Two named tables with one name are refused, naming the table, so give one of them a different "name".

"delimiter" is told rather than sniffed. Half of Europe writes a CSV with semicolons because the comma is its decimal point, and read with commas such a file comes back as one column called id;name holding 1;ada: no error, no missing rows, and nothing to act on. Guessing from the look of a file is wrong on one in fifty and silent about which, so it is a setting. One character, and only on a "csv" table.

Paths resolve against the configuration file, so a config and its data move together. name is optional for files and defaults to the stem; an HTTP source must be named, because a URL has no obvious table name.

An HTTP source fetches JSON and shapes it with the same rules as a JSON file. Fifty-four public endpoints were surveyed to decide what those rules are; see docs/api-shapes.md. scripts/api_survey.py grades the whole pipeline against 282 of them, through to the SQL answers agreeing with the data they came from.

These are the only keys read here, and a key that is not one of them is refused rather than ignored, naming the nearest one that is: an option written beside "http" instead of inside it leaves a config that looks right and a source that behaves as though the line were absent.

A value written as text is read as a number only when nothing it spelled is lost by doing so. CSV and XML have no types at all, so a number there can only arrive as text and has to be recognised; JSON has types per value, and a string of digits is a string the source chose to write. One rule serves both: believe the source unless the conversion is exact. Measured over 239 public API responses, the old rule typed 791 columns numeric from text and 377 of them lost something. Coinbase quotes rates to 19 significant digits as JSON strings and a float holds 17; ipapi writes utc_offset as "-0700", and -700 is a different thing.

records defaults to auto, which scores the readings of the document and refuses a weak winner rather than guessing. Naive detection is the trap here: over 85 public endpoints, a detector that simply looked for an array found one for 84 of them and was frequently wrong, serving paging links, a nested field, or a rejected request's errors list as the table. Scored detection is right on all 40 endpoints whose correct answer was written down first, and costs 0.01 to 0.06 ms. Name records explicitly to turn detection off.

The hard case is a map of same-typed scalars. Frankfurter answers with 29 numbers under USD, GBP and SEK; sunrise-sunset answers with 10 strings under sunrise, solar_noon and day_length. Both are a map of scalars, so neither the value types nor the number of keys separates them. One rule does: a key that identifies a row was made by whatever makes that domain, so every key in the map shares a shape, and a key that names a field was chosen by a person writing a schema, so they share nothing but being words.

Agreement only counts once there are enough keys for it to be unlikely. Two keys sharing a shape is a coincidence that happens constantly: across those 116 responses there were 44 distinct two-key objects, 11 of them with keys of one shape, and not one was rows. lat and lng are three lower-case letters each; so are sha and url, svg and png. From three keys up, the same corpus had 7 agreements and every one was rows.

Discovering an API

discover takes the base of an API and works out what is on it. Three routes, tried in that order:

1. A description document. OpenAPI at openapi.json, swagger.json, v3/api-docs, swagger/v1/swagger.json, .well-known/openapi.json or api-docs, under the base and then at the origin. Every GET path without a parameter in it becomes a candidate.
2. A link index. Many APIs answer their own root with a map of name to URL. HAL _links and JSON:API links count too.
3. Conventional names, only when the first two found nothing at all. Some real APIs publish neither an index nor a description, and asking for the names such an API probably uses is the move left.

The crawl runs even when a description was found, because a description is authoritative about what it names and silent about what it omits.

An entry may also be the base URL alone, as a string.

Links inside row data are not followed. A collection of 20 characters holds 20 links to 20 individual characters, and following them produces 20 more one-row tables named after the one you already had. URI templates are not fetched: https://api.github.com/repos/{owner}/{repo} is an invitation to substitute, not an address.

Pagination

A collection served a page at a time is read whole. Discovery works out how, from the response rather than from a list of API names: a next link where there is one (next, info.next, links.next, _links.next.href), and otherwise the position the envelope reports about itself. dummyjson answers with skip: 0, limit: 30 beside a total of 194, GBIF with offset and limit, Algolia with page. Each of them is echoing back the parameter it was given, which is what makes advancing it general rather than a guess.

Pages are then fetched in parallel. Two consecutive links show which parameter moves and by how much, so the rest can be written down instead of asked for one at a time; measured against a server held at 40 ms, 200 rows in 20 pages went from 0.99s to 0.33s. The end of a collection is still the API saying so, not a guess from a row count: an empty page, a repeated one, or one with no next link of its own.

If an API ignores the parameter and answers with page one every time, the second page is identical to the first and the read stops there rather than serving twenty copies of it. max_pages (50) and max_rows (100,000) bound the rest. Stopping at max_pages with more available serves what was read and logs it, because a partial collection served silently is the one failure a person querying the table cannot see. Passing max_rows refuses the source rather than serve part of it, naming max_rows.

Requests to one host are capped at four at a time. Sources load in parallel and each may be paging, so without that a catalog of fifty PokeAPI tables opens several hundred connections to one server: enough, measured, for an API to start refusing.

Table lists and queries

INFORMATION_SCHEMA wants to know what exists and what its columns are, and for that a source's first page is enough. Reading every page of every source to answer it took 23 seconds on a catalog of 65 discovered tables, and reading first pages only took under a second. The startup warm reads first pages only, and so does finding the tables made from arrays inside a source's rows (see Nesting): a ten-page source takes one request to warm. A query reads the whole table, child tables included.

The one thing the two can differ on is a column type, since types are inferred from the values present and a later page can hold a float in a column whose first page was whole numbers. The query is unaffected: it infers over everything it read.

XML and HTML

Neither gets its own pipeline. Both become lists and dicts and then go through exactly the same detection, flattening and typing as JSON, so an RSS feed and a JSON envelope with rows under items end up as the same table.

{
  "tables": [
    { "name": "headlines", "http": "https://feeds.bbci.co.uk/news/rss.xml" },
    { "name": "elements",
      "http": "https://en.wikipedia.org/wiki/List_of_chemical_elements" },
    { "name": "catalog", "xml": "data/catalog.xml" },
    { "name": "page", "html": "data/page.html" }
  ]
}

The format is sniffed from the first bytes. Content-Type is wrong often enough to matter, and a feed served as text/html would be unreadable if the header were believed. Set "format" to json, xml or html to say outright.

CSV is the exception: it announces itself nowhere in its bytes, so it is asked for rather than sniffed. "format": "csv" reads a response whose first line is its header, which is what an open data portal serves, and from there it is the same table as any other source. A byte order mark is stripped, a quoted field may hold a comma, and a line with the wrong number of fields is refused with its line number rather than padded with NULLs.

{ "name": "passengers", "format": "csv",
  "http": "https://example.org/titanic.csv" }

In XML, attributes become @-prefixed columns so <link href="..."> and <link><href> do not collide, a repeated tag becomes rows, and namespaces are stripped: {http://www.w3.org/2005/Atom}title cannot be typed into a query. A DOCTYPE is refused rather than parsed, because a DTD can define entities that expand a small document into gigabytes inside the parser.

From HTML you get every <table> on the page, keyed by its caption or its position, plus any JSON-LD the page publishes. Layout tables with a single row are dropped, <style> blocks inside cells are ignored, and the header is the widest-by-distinct-values row in the run of <th> rows at the top, which is what separates a real header both from a colspan sub-header below it and from a spanning title above it. Page layout is not scraped: a <table> is the one thing on a page that is already a table.

Excel and Access

Both hold more than one table, so both produce more than one. A workbook makes a table per sheet and one per table drawn on a sheet, and a database makes one per table and per saved query, each named after itself:

{
  "tables": [
    { "excel": "data/budget.xlsx" },
    { "excel": "data/budget.xlsx", "sheet": "Q1", "name": "first_quarter" },
    { "excel": "data/budget.xlsx", "table": "Headcount" },
    { "excel": "data/budget.xlsx", "sheet": ["Q1", "Q2"] },
    { "access": "data/club.accdb" },
    { "access": "data/club.accdb", "table": ["Members", "Rooms"] }
  ]
}

Point at a file and you get everything in it. "sheet" and "table" narrow that to what they name, one name or a list of them, and naming tables alone serves no sheets, which is how to say "the tables, not the tabs they sit on". "name" renames what is left, and is refused where more than one thing is, because there is one name and four tables and three of them would end up called something invented.

What Excel calls a table, and the object model calls a ListObject, is a range somebody drew and named. It is stored in a part of its own rather than in the sheet, so it carries three things a sheet cannot: a name, its own column names, and where it stops. A tab and a table on it are both served, and where the tab holds only that table the two are the same rows under two names.

A tab is not served beside its tables when the file says its own reading would be wrong, and it says so four ways. A table named after its own tab replaces it, because naming the table after the sheet is what people do and one name cannot mean two tables. A tab carrying more than one table has more than one header row, so no reading of it as a single table is right: serving one would put the second table's headings in as a record and drag its numbers over to text with them, a wrong answer with nothing on the wire to show it. A table ending in a totals row would hand the tab a record called Total, counted by every count and added into every sum. And a table with no header row leaves the tab nothing to name its columns with, so the tab read whole would take its first record for the headings. The log says which reason applied, and naming the tab with "sheet" serves it regardless.

A workbook is read directly. An .xlsx is a zip of XML and the standard library opens both, so nothing has to be installed to read one. That matters more than it sounds: the usual way to read a workbook on Windows is the Access database engine, which is a separate download that installs in one bit width and refuses to load into a process of the other, and a bridge whose promise is "point it at a file" cannot begin by asking for a driver. Neither reader below touches it, or COM, or ODBC. An .xls saved by an old Excel is a different format and is refused with a message saying so.

The first row of a sheet names the columns. That is a rule rather than a reading: a sheet with a title above its headings gets the title, which is visible at once and fixed by drawing a table over the headings and the rows under them, where guessing which row looked most like headings would be wrong occasionally and silently. A table is the answer to that layout, because it says where the headings are instead of leaving it to be inferred.

A tab that cannot be read as one table is left out, and the rest of the workbook is served. That covers a value to the right of the last heading, which nothing could name, a heading that cannot be a column name, and more columns than a result can carry. The log names the tab and the cell, and says what was served in its place: a tab that carries a table is served as that table. Name the tab with "sheet" and the reason comes back as an error instead, and a workbook left with nothing to serve is refused with every reason. Only the layout is forgiven. A cell that cannot be read at all, or a file that is not a workbook, still refuses the whole file.

A cell that is empty is NULL, a row that is not in the file is not a row, a row of nothing but #DIV/0! is a row of NULLs, and a formula arrives as the value Excel last worked out for it.

Dates are the part of the format worth knowing about. Excel stores a date as a number of days and the only thing that makes 45306 a date rather than the number 45306 is the number format its style points at, so the styles are read to find out. Day 60 is Excel's 29th of February 1900, a day that did not happen, and is refused rather than served as some other date; a workbook saved by Excel for Mac before 2011 counts from 1904 and is read against that epoch, because a date read against the wrong one is out by four years and a day with nothing looking wrong.

An Access database is read in pure Python too. The .mdb and .accdb formats are undocumented and page-structured, so the reading is not done here: pyOpenVBA implements the Jet storage engine and this maps what it hands back onto the columns it serves. It is a dependency with no dependencies of its own and the same Python floor as this, so nothing else comes with it.

The alternative was the database engine Microsoft ships, through COM or ODBC, and the reasons against it are the reasons against it for the workbook: a separate download, installed in one bit width and refusing to load into a process of the other, and absent from Linux entirely. Reading the file instead means an Access database is served on any machine the rest of this runs on, that the suite tests it on every one of them rather than skipping where an engine is missing, and that the single-file executable needs nothing installed beside it.

The file is read once into memory and never written. That is stronger than opening it read-only: there is no handle held, no lock file beside it, and somebody can have the same database open in Access while it is being served.

Saved queries are served alongside the tables, and run where they were written. An Access query says IIf and Nz and joins its strings with &, none of which is what a client sends this, so the query is answered in Access SQL and its rows arrive here already worked out. Only the ones that read: an update or a delete query is a statement rather than a table, and running one to find out what it returns would change the file. A query has no stored schema, so its columns are read off its rows, and one that answers nothing has no columns to be described by and is passed over.

Types come from the database, because unlike a CSV it has some. A text column stays text where every value in it happens to be digits, which inference alone gets wrong, and Access's two-byte Integer stays two bytes. Jet declares a text length in bytes, two to a character, so TEXT(50) arrives as 100 and taking the number at face value would give a column half the width it should be. Where there is no type here that is theirs the values decide instead, which covers Yes/No, currency and the GUID: a currency column becomes a float where every value survives one exactly and text where one of them would lose digits, which is the rule a number arriving as text already goes through. A column holding an OLE Object is refused by name, because there is no column type here that carries an embedded file and answering NULL would be a quieter lie.

scripts/office_probe.py reads a workbook or a database with these readers and prints what came out: the tables, the type each column was given, and the first few rows. For a file that will not serve, or serves something unexpected; if it prints what you expect and a client does not see it, the problem is past the reader.

Nesting

A row cannot hold a list. Over 116 public API responses, a third of the tables built from them had an array in every row: a character has episodes, a cart has products, a recipe has ingredients. Serving those as JSON text makes a column nobody can query, and dropping them loses the data.

An array inside an HTTP or discovered source's rows becomes a table of its own, named parent_column, with the parent's key beside every element. An array inside a JSON, XML or HTML file stays a column of JSON text.

characters              20 rows   id, name, status, species
characters_episode     242 rows   characters_id, episode_index, value

SELECT c.name, COUNT(*) AS episodes
FROM characters c JOIN characters_episode e ON e.characters_id = c.id
GROUP BY c.name ORDER BY episodes DESC

It recurses. A cart holds products and a product holds reviews, so there is a carts_products and a carts_products_reviews, and every level carries the identity of the levels above it: a review row has carts_id, products_index and reviews_index, so it can be joined straight back to the cart as well as to the product. Four levels deep and 64 tables per source are the bounds; the corpus produced 123 child tables from 114 parents, 67 of them two or more levels down.

The key is found by testing, not by naming: the first column whose values are present in every row and never repeat identifies those rows, which is what a key is. A column called id that repeats is not one, and a slug that does not repeat is. It is carried under the parent table's name, because an element usually has an id of its own and writing both as id would lose the parent and make the obvious join match the wrong thing.

Nothing that became a table is also left behind as JSON text. Set "expand": false on a source to keep the old behaviour.

Depth was never the problem. Nothing in that corpus nested deeper than the flattener already goes.

Credentials

{ "auth": { "bearer": "${GITHUB_TOKEN}" } }
{ "auth": { "header": "X-API-Key", "value": "${WEATHER_KEY}" } }
{ "auth": { "header": "Authorization", "prefix": "Token", "value": "${PAT}" } }
{ "auth": { "query": "api_key", "value": "${NASA_KEY}" } }
{ "auth": { "basic": { "username": "someone", "password": "${PASSWORD}" } } }

${VAR} and ${env:VAR} read the environment. A config file gets committed and a token in a committed file is a leaked token, so the config holds the name and the value stays outside it. A literal is accepted for local testing. The secret never reaches a repr, a log line, or an error message: a failure names the variable it came from, not what was in it.

The credential, and the headers beside it, go only to the servers the source's url names, matched on scheme, host and port. A next link or a page is whatever the API's answer says, so an answer could otherwise point the next request at another host and be handed the token. A link to another server is still followed, without them, and the log says so once per host. A redirect to another server drops them too. An API that genuinely serves its pages from another host names it in "credential_hosts", over https only:

{ "auth": { "bearer": "${TOKEN}" }, "credential_hosts": ["cdn.example.com"] }

Several URLs are raced, and the first successful one wins: a replica that fails fast should not beat one that succeeds slowly. Sources load in parallel and the server warms them at startup, so the first client waits for none of it. Once something is cached, an expiry serves the previous answer immediately and refreshes behind it.

A source that cannot be reached is listed in the catalog with no columns rather than failing the whole table list, and selecting from it reports why it could not be loaded instead of claiming the table does not exist.

Column types

A column's type is inferred from every value in it, not per row, because COLMETADATA declares it once and every row is encoded against that declaration. One non-integer drops the whole column to float, one non-number drops it to text. Empty CSV cells are NULL. JSON objects are unioned across records, so a missing key gives NULL rather than shifting the row. A nested object becomes dotted columns, as address.city. An array is served as its JSON text, or as a table of its own from an HTTP source (see Nesting). With "flatten": false a nested value is refused rather than stringified.

Logins

Windows Authentication is what a client uses by default and needs nothing configured. The login carries a SPNEGO token and SSPI's AcceptSecurityContext validates it against the local account database or the domain, so no Windows password reaches this bridge. A username and a password is the other way in, checked here, and it is off until a configuration names the accounts that may use it:

{
  "logins": [
    { "user": "reader", "password": "${env:BRIDGE_READER_PW}" },
    { "user": "excel",  "password_hash": "pbkdf2_sha256$210000$c2FsdA==$..." }
  ],
  "tables": [
    { "name": "people", "json": "data/people.json" }
  ]
}

This is the opposite direction from "Credentials" above. That one is this bridge proving itself to an API it reads from; this one is a client proving itself to this bridge.

A bridge with no "logins" refuses every username and password there is. What a client sees is Login failed for user 'x'. with message number 18456, which is the number and the words a real server gives, measured against SQL Server 2025. An unknown name and a wrong password get that same answer and take the same time to get it, so a refusal never says which of the two it was, and the server log names the login that was refused and never what was offered for it.

Write "password_hash" rather than "password" anywhere the file will be committed. pysqlbridge --hash-password asks for a password and prints the hash to paste in; it asks rather than taking an argument because a password written on a command line is in the shell's history and, while it runs, in the process list. A "password" is read from the environment through ${env:NAME}, and a variable that is not set stops the server at startup naming the variable, rather than refusing logins later for a reason nobody can see. Usernames are matched without regard to case, passwords exactly.

None of this makes the connection private on its own; see the note on the password field in the protocol notes below.

The SQL it answers

Measured rather than chosen: thirty queries a client or a person would plausibly send were run through the whole stack, and the sixteen that were refused set the order of work. Twenty-nine now answer.

SELECT p.name, COUNT(*) AS posts
FROM people p
JOIN posts o ON o.userId = p.id
WHERE p.name LIKE 'C%'
GROUP BY p.name
HAVING COUNT(*) > 5
ORDER BY posts DESC
OFFSET 0 ROWS FETCH NEXT 10 ROWS ONLY

Nothing that writes is supported, apart from the temporary tables a connection builds for itself: a client makes one, or has a SELECT ... INTO make it out of the answer, fills it naming the columns or taking them in order, changes it with UPDATE #t SET ... WHERE ..., DELETE FROM #t WHERE ... and TRUNCATE TABLE #t, reads it back and drops it, and nothing a source holds is touched. Those three leave @@ROWCOUNT and give a missing table or column the numbers a real server does, and an UPDATE that fails part way leaves every row as it was. A form beyond them, with a FROM, a TOP, an OUTPUT or a variable being set, is refused by name rather than run.

A value put into a column becomes the column's declared type the way a cast would make it, measured case by case: '7' into an int is 7, 5.7 is 5, 'q' is msg 245, 300 into a tinyint is 220, and a datetime column holds moments rather than the text it was given. Text is the exception: a cast cuts it short, and a column refuses it with 2628, unless all that would be cut is spaces. An INSERT keeps all its rows or none, so one row that will not fit leaves the table as it was. A table a SELECT ... INTO made converts to the types of the answer that made it, except text, which it keeps as given: a real server holds that to the answer's size, and here a text column is sized by the values read rather than declared. And a 245 about a value that reached the table through a read names it nvarchar where a real server says varchar, because every column this serves is nvarchar; the number is the same. An INSERT, UPDATE, DELETE, MERGE, TRUNCATE, DROP or ALTER naming anything else is refused and says so, and so is a GRANT, REVOKE or DENY: there are no permissions here to change and every source is read-only for everyone who can reach it. Passing any of them over would report that it worked, and a person told their DELETE succeeded has been told something untrue about their data. The statements a client sends to open a session, SET and USE and the rest, are still passed over.

Transactions are accepted and counted. BEGIN TRANSACTION, COMMIT, ROLLBACK and SAVE TRANSACTION move @@TRANCOUNT, which is kept per connection: nested begins count up, a rollback to a savepoint releases that savepoint and every one marked after it, and a commit or rollback with nothing open is refused with SQL Server's own number and words. A client's transaction API reaches the same count as the statements do, because it is answered by them: a .NET BeginTransaction() arrives as its own packet rather than as SQL, and is turned into BEGIN TRANSACTION here. What a transaction does not do is undo anything, because nothing here is written. A rollback restores the count, not the rows a batch put in a temporary table, and a distributed transaction is refused rather than joined, whether a client's transaction API asks for one or it is written as BEGIN DISTRIBUTED TRANSACTION.

A client that pools its connections asks for the session to be reset on the first message it sends over one it has taken back out, and that is answered now. Everything the last user of that connection left behind goes with it: the temp tables it made, the transaction it left open, and the SET options it changed. Measured on SQL Server 2025 by closing a pooled connection and opening another with the pool held to one, so the same physical connection comes back. Until this the request was read and ignored, so one user's scratch tables and settings were still there for whoever had the connection next, which is the wrong answer and somebody else's data. The other form of the request, which keeps the transaction and clears the rest, is answered too; a client only sends that one for a distributed transaction, and those are refused here.

A batch does not always stop at its first error, because a real server does not. Which happens depends on the error, and each number was measured on its own: after 3902, 8134, 2812 and ten others only the failing statement ends and the rest of the batch runs, with the error arriving among the answers rather than in place of them; after 208, 245, 628 and five others the batch stops, and what the statements before it answered is still returned. A client reads results, then the error, then more results, in the order they happened. An error outside both sets ends the batch with nothing kept, which is what a real server does with one it cannot compile. An error raised inside an EXEC of text ends the text and no more, and the batch that ran it carries on; 241, 245, 281, 628, 3623, 8114 and 8169 are the ones that reach out of the text and end that batch too.

A batch that cannot be T-SQL at all runs none of its statements, because a real server compiles the whole batch before it runs any of it. Measured: CREATE TABLE #t (a int); SELECT * FROM is msg 102 and leaves no #t, where this made the table and then failed, and a batch cut off after DECLARE @p, or a SET with a SELECT after it, used to answer as though it had worked. The text is checked before anything is parsed, and only where the text alone settles it. A quote or a bracketed name left open is 105 and then 102 near what it left open, a comment left open is 113 (comments nest), and a keyword such as FROM where a name or a value belongs is 156 near it. Text that stops after WHERE, a comma, an open bracket or a BEGIN with no END is 102 near its last word, quoted as it was written. A SET that begins its statement and stops short is 156 near SET, and one belonging to an UPDATE is 102. The first error is the one a client raises, and a second goes out behind it the way a real server sends one. Every rule and every word treated as wanting more was measured, and each construct that looked like it might trip one, such as IS DISTINCT FROM, FOR SYSTEM_TIME ALL WHERE or a cursor's trailing FOR UPDATE, was parsed by a real server under SET PARSEONLY ON and is left alone. A real server that has found one error reads on and can report another from later in the batch. This reports the first, and the lexer's error behind it where the text ends inside a quote or a comment.

A value written where a condition belongs is msg 4145, because T-SQL has no boolean to read one as: WHERE score, WHERE 1 and WHERE CAST(1 AS bit) alike, and ON, HAVING, IF and a CASE's WHEN the same way. IIF takes a condition too, so IIF(1, 'a', 'b') is refused where this answered 'a'; a real server reads an IIF as the CASE it stands for, and the message names the bracket its arguments open with. A batch cut short inside a condition is 4145 as well, wherever it stops afterwards, so WHERE rank and WHERE rank ORDER both are while WHERE rank = 1 is 102 for the statement being unfinished. What decides it is the condition parser itself, asked about the condition alone and only where the text was going to be refused anyway.

A window function's argument may be a subquery, which a real server takes and says so in the message that refuses a column there: NTILE((SELECT 2)), LAG(score, (SELECT 1)), FIRST_VALUE((SELECT 5)) and SUM((SELECT 1)) OVER (...) all answer, and all were refused here because the arguments were never lifted the way a select list's are. An aggregate that reduces the rows is the other way about: SUM((SELECT 2)), SUM(SUM(score)) and SUM(score + (SELECT 2)) are msg 130, settled while compiling, where this answered them.

An expression with nothing in it to take a type from is refused for the same reason. A CASE whose every result is the word NULL is msg 8133, IIF with both results NULL the same, COALESCE with every argument one is 4127, and NULLIF whose first argument is one is 4151. A NULL that carries a type is not the constant, so CAST(NULL AS int), a declared variable and a column are all fine, and one result with a type settles the rest. All four were answered as NULL here.

A SELECT with no FROM that reads a column says so in SQL Server's words rather than this project's: 207, "Invalid column name", naming the first column it cannot find, 4104 where the column is qualified, 263 for a star and 107 for a star with a prefix, all measured. The functions T-SQL writes with no brackets, CURRENT_TIMESTAMP, SYSTEM_USER, USER, CURRENT_USER and SESSION_USER, are values rather than columns there and everywhere else, unless bracketed as [user]; they were refused.

A name qualified by a table the query does not read is msg 4104, "The multi-part identifier could not be bound", and a star with such a prefix is 107: the qualifier has to match a table or an alias the FROM names, an alias hides the table's own name, and a qualifier of more parts is bound by its last, so dbo.people.id is people.id. Each branch of a union binds its own. A qualified name used to fall back to its bare column, so a query naming a table it had renamed answered rows.

An aggregate with no FROM reduces the one row a select with no FROM has, so SELECT COUNT(*) is 1 and SELECT COUNT(*) WHERE 1 = 0 is nought, because the WHERE takes that row away. HAVING works over it the same way, and a GROUP BY there is msg 164: there is no column to group by.

A variable nothing declared is settled at the same time and the same way: msg 137, or 1087 where a table goes, and none of the batch runs. It used to read as null, so a variable name spelt wrong came back as an empty answer instead of an error. Declaration follows the text rather than what runs, as it does on a real server, so IF 1 = 0 BEGIN DECLARE @x int END; SELECT @x AS v answers. A DECLARE's own variables are not declared inside it, so DECLARE @a int = 1, @b int = @a is 137 on @a. Each statement that reads one reports it, an IF's condition and the statement it guards separately. The values a client sends with a parameterised statement declare their names, and so does a procedure's header. An EXEC's @p = 1 names the procedure's parameter rather than a variable of the batch, while EXEC @rc = p reads @rc. All of this was measured. scripts/replay.py supplies a parameter the log did not keep as null, because the log records a parameterised statement without its values, and it says how many queries needed that.

A SELECT that assigns from a table assigns once for every row, and each row reads what the one before it left, so SELECT @s = @s + name + ',' FROM people ORDER BY id builds the list up. It used to read every row with the value from before the statement and keep the last row's answer, which gave only the last name. Several variables in one SELECT are assigned left to right, so SELECT @a = 1, @b = @a + 1 leaves @b at 2. A TOP applies before any row assigns, a WHERE that keeps no row leaves every variable as it was, and an assignment made before a failure part way through is kept. SET @a += 2 and the other compound operators are the variable with the operator applied to all of what follows, and DECLARE @a int = 1, @b int = 2 gives each its value in turn. Both used to be wrong: the first was passed over as a SET option and left @a alone, and the second refused, or where the first variable had no value, left the rest null. A SELECT that assigns and reads in one list is msg 141, settled while compiling. The one form still refused is a GROUP BY whose items read a variable they assign, because each group would need what the group before it assigned.

Text an EXEC runs is a batch of its own, in a scope of its own. It sees none of the variables of the batch that ran it and leaves none behind; what it is told is the parameters sp_executesql's declarations name, given by name or by place and held as the type declared for each; and a parameter marked OUTPUT goes back into the variable it came from once the text has run. The text itself may be worked out rather than written out, so EXEC(@sql) and EXEC('SELECT ' + @what) run what they build. What it is handed is checked the way a real server checks it: a parameter nothing supplied is msg 8178, one argument too many 8144, a value after a @name = value 119, OUTPUT on a constant 179, OUTPUT for a parameter not declared as one 8162, and a statement that is not nvarchar 214.

Every message that names the text a conversion was given says which text it was. A real server types 'ab' as varchar and N'ab' as nvarchar, so CAST('ab' AS int) says varchar and CAST(N'ab' AS int) says nvarchar, and so do the arithmetic, the casts that name both types and the overflow. A cast to varchar, char or text makes narrow text and the next message about it says so. Every column this server serves is nvarchar and is named that way. All of it used to say nvarchar.

A batch that is nothing but a DECLARE or a SET still runs. Measured: DECLARE @x int = 'abc' on its own is msg 245, DECLARE @x decimal(3,1) = 12345.6 is 8115 and DECLARE @x int = 5/0 is 8134, where this answered nothing at all because no statement in the batch read anything.

A variable holds the type it was declared with. Every value given to it, by DECLARE, SET or SELECT, is converted the way a cast converts, so an int given 5.7 holds 5, a varchar(3) given a name holds its first three letters, a varchar with no size holds one, and a decimal(5,2) rounds. Each row of a SELECT that assigns is held as the type before the next row reads it, so a running sum into an int drops the fractions as it goes, the way a real server's does. A value too big for the type is refused with the error the cast gives and the variable keeps what it had. A variable's name is matched without regard to case.

A read that fails while working out a row, a divide by zero or a conversion that will not go, has already had its column shape sent by the time it fails, so a real server puts an empty result set naming those columns in front of the error. That shape is worked out from the select list without running the query and goes out ahead of the error here too, whether the read failed on its own or among a batch's answers. A read that fails while binding, an invalid object or an unknown column, sends none, because it never got that far: the fourteen numbers that keep the shape are the ones that come from evaluating a row, measured one at a time.

RAISERROR raises the error a client asked for: message number 50000 with the text, the severity and the state it was given, catchable, readable through @@ERROR, and the batch carries on past it the way a real server's does. It was passed over in silence before, so a client that raised an error deliberately was told everything had gone well. Severity 10 and under is a message rather than an error on a real server, and there is no way to send one from here, so those go over the way PRINT does.

THROW raises one too, and differs in what the batch does next: a real server stops there, where it carries on past a RAISERROR. What the statements before it answered is still returned, the number is the one it was given rather than 50000, the severity is always 16, and the state is its third argument. A bare THROW inside a CATCH re-raises what was caught, whole, and ends the batch even where that error on its own would not have, which is why the error carries a mark and the number is not asked. A number outside 50000 to 2147483647 is refused with 35100 as a real server refuses it, and a bare THROW with no CATCH around it is 10704, which a real server settles while compiling, so nothing in the batch answers. Inside an EXEC of text it ends the batch that ran the text as well, where a RAISERROR in one does not.

SET XACT_ABORT ON changes where a batch stops rather than what it keeps. Measured with the setting on, one number at a time: twelve of the thirteen errors a batch would otherwise run on past end it instead, and what the statements before them answered is still returned. The exception is 3701, dropping a table that is not there, which is the one of them a real server reports at severity 11 rather than 16. An error a client asked for is not the setting's to change: a RAISERROR still lets the batch run on and a THROW still ends it. A CATCH still catches. The setting belongs to the connection rather than the batch and survives into the next one, and SET XACT_ABORT OFF puts it back.

A batch that stops at an error leaves no transaction open, and takes every nesting level at once, so @@TRANCOUNT reads nought afterwards rather than what it read before the failure. A batch that carries on past its error keeps the transaction. Two kinds of error do not roll one back: a compile error, 102, 105, 113 or 156, each measured with the setting off and on, because nothing compiled and so nothing ran to undo; and 208 with SET XACT_ABORT off, which is the one number the setting changes the answer for. A refusal of this server's own keeps the transaction too; a real server has no counterpart for one, so that is chosen rather than measured, and it is the safer way round, because a client committing a transaction this had thrown away would read 3902.

@@ERROR reads the number of the last statement that failed, and nought where it worked, so IF @@ERROR <> 0 does what a client means by it. It answered NULL before, and nothing is ever unequal to NULL, so that check quietly never fired. A DECLARE with no value leaves it standing, the same exception @@ROWCOUNT makes for one; an IF and an EXEC of text leave it to the statements they ran; a TRY clears it once its CATCH is done, and a CATCH reads the number that sent it there. What a TRY answered before it failed is kept too, because a real server has already sent those rows by the time it reaches the failure.

A WHILE runs its body until its condition is not true, and asks the condition the way an IF does, so a loop that ends there reads nought from both @@ROWCOUNT and @@ERROR, while one ended by a BREAK reads what its body last left. BREAK and CONTINUE reach the loop from however deep in its blocks, IFs and TRYs they stand. Outside any loop they are msg 135 and 136, settled while compiling, and the text an EXEC runs is outside the loop around the EXEC, all of it measured. It ran the body once before, so WHILE @i < 3 SET @i += 1 left @i at 1 and reported success. A real server will loop for as long as it is asked to; this stops a loop at its 100,000th turn and says so, rather than holding a connection forever.

Inside a CATCH, ERROR_NUMBER(), ERROR_MESSAGE(), ERROR_SEVERITY() and ERROR_STATE() answer what sent it there, and go on answering it for the whole block where @@ERROR does not: a statement running inside the CATCH clears @@ERROR and leaves these alone. Outside a CATCH every one of them is NULL. ERROR_PROCEDURE() is always NULL because nothing here runs in a procedure, and ERROR_LINE() is as well: a real server answers the line within the batch, nothing here counts lines, and a number would be invented.

Windows, casts, dates and UNION

A window function is worked out over the rows the WHERE kept, before the sort and before TOP, and answers once per row. Where a query does not name a frame it gets the one SQL Server uses: the whole partition where the OVER clause says no order, and everything up to and including this row's ties where it does. That is why SUM(x) OVER (ORDER BY id) is a running total and SUM(x) OVER (ORDER BY team) is not. A query may name one instead, as ROWS or RANGE, which is how LAST_VALUE is told to look at the whole partition rather than stopping at this row. A window in the ORDER BY rather than named in the select list, and a window beside a GROUP BY, are each refused by name rather than answered differently. NTILE's count may be a variable and must be an integer above nought. LAG and LEAD read their offset and their default from the row asking, so LAG(x, @n) and LAG(x, rank) both work, a null offset gives the default, and a negative one is refused.

A cast to an integer type is held to the range of that type, so CAST(300 AS tinyint) is an error rather than 300, and TRY_CAST and TRY_CONVERT answer NULL wherever CAST refuses. That pair matters more here than on a real server: a source read off a CSV or an API holds whatever it holds, and one value that will not convert should not cost the answer. A cast to text keeps as many characters as its size says: thirty when it gives none, 128 for sysname, and all of them for nvarchar(max), varchar(max), text and ntext. A cast to decimal(p,s) rounds to its places, half away from nought, and refuses a whole part too long for it; money keeps four places; a date keeps the day and drops the time. A bit becomes the text 1 or 0, and the words true and false become bits. A float written out as text keeps six significant digits and goes scientific where they will not reach, so a third is 0.333333 and a million is 1e+006; a decimal keeps all of its places.

The date functions are measured the same way, and most of what they do is not guessable. DATEDIFF counts the boundaries between two moments rather than the time between them, so a minute either side of midnight is one day and a whole day inside one date is none. DATEADD holds a month back rather than letting it spill, so a month after the 31st of January is the 28th of February. Weeks start on Sunday and week one is whichever week holds the 1st of January, so 2026 runs to week 53. GETDATE() is taken once for the whole statement, because a filter comparing each row against its own slightly later now would keep different rows for no reason.

Where a UNION puts two columns together, the result gets one type, chosen across every branch by SQL Server's data type precedence and measured against it pair by pair. A union of an integer column and a float one is float and keeps the fraction rather than truncating it to the first branch's type, and a value that will not convert is refused with the number and wording a real server refuses it with.

Measured against SQL Server 2025

The semantics are not chosen, they are compared. scripts/differential.py writes a fixture twice, once as JSON for this and once as INSERT statements for SQL Server, and scripts/differential.ps1 runs 1,243 queries against both and reports where the answers differ, by value, by the kind of each column and by its heading. The heading was the last added: an unaliased window function was headed by its own name here and by nothing there, and every query that read one agreed until the names were compared. Where both refuse, it compares the number as well as the words: a client shows it, and a divide by zero reported as msg 208, invalid object name, sends whoever reads it looking for a table that was never the problem. The rows sit on the edges rather than the middle: NULL in every position that treats it specially, text differing only in case, an empty string, a zero, a negative, and a key that matches nothing. A third table holds values far enough apart that adding them in one order and the other give different floats, which is how the order the arithmetic runs in gets compared at all. A fourth is written as a workbook rather than as JSON, because a workbook is the source that can hand over a real moment, and there is no other way to put a datetime column on this side of the comparison.

Adding the workbook table found four things, every one of them wrong here. WHERE hired = '2024-01-15' answered nothing, because Python calls a datetime and a string unequal rather than refusing to compare them, so the row simply did not match; WHERE hired > '2024-01-15' was right only by the accident of an ISO date sorting the same way as its own spelling, and stopped being right at '2024-6-1', which a real server reads and this refused. WHERE hired IN ('2024-01-15') matched nothing for the same reason as the first. CAST(hired AS int) refused a conversion a real server answers with 45304. None of the four was reachable before, because no source produced a datetime column: JSON has no dates and a CSV's are text.

-AsRpc sends the same battery the other way. A parameterised statement reaches a server as an RPC call to sp_executesql rather than as a batch, which is a different path through this one, and it is the path a client uses: SSMS sends 50 of its 93 queries that way. Adding a parameter to each command makes the driver do it, and the statement reads the parameter in a check of its own so that the switch cannot quietly test nothing. Both differential.ps1 -AsRpc and batches.ps1 -AsRpc answer identically to a real server, which is what says the two paths agree.

A query named mine-only- is one this answers where a real server refuses, on purpose and with the reason written beside it. The harness reports those separately rather than counting them as agreement, and complains if one stops being answered here or starts being answered there, so that a divergence nobody decided on cannot hide among the ones somebody did.

scripts/replay.py asks a different question. It reads a log written by --log while a real client was connected and answers every query in it again, so what comes back is not whether an answer is right but whether there is one, over what a client actually sends rather than what anyone thought to write down. SSMS sends 93 distinct queries before it will draw a tree, all of which are answered, and any that refuse are the work list.

Thirteen differences turned up that way, every one of them wrong here:

COUNT(DISTINCT x) and aggregates over an expression came out of the same run, as things a real server answers and this refused.

scripts/truncations.ps1 asks about text a real server refuses. It sends every query in the battery cut short at each word, 8,290 prefixes, to both servers, and sorts each prefix into one of four outcomes. Both refusing with the same number is agreement. Both refusing with different numbers is a client shown the wrong message. A real server answering what this refuses is usually something not supported here. This answering what a real server refuses is the worst of the four, because it is a truncated query given an answer that nothing about it admits. Its first run found 175 of those, among them a CREATE TABLE cut off after a comma that made a table, and 5,447 prefixes refused with the wrong number, nearly all of them syntax errors reported as 50000. Checking a batch for syntax before running any of it brought those to 18 and 975; settling the rest of the syntax, the binding and the types brought both to none and 73. Nothing here now answers a prefix a real server refuses, nothing refuses one a real server answers, and the 1,108 both answer agree on every value. The 78 that differ are error numbers alone, and what is left of them is a real server finding something earlier in the text than this does. Two are a WHILE cut off inside its condition, which this refuses as 50000 for the condition having no end. Twelve are a query holding an IIF whose condition is a value, which is 4145 there and a syntax error here because the text stops before the select list is read; eleven are a column that cannot be bound, which a real server reports before the 130 or the 8133 the same statement also earns.

One divergence is known and open. A literal written with a decimal point is decimal on a real server and a float here, so arithmetic over one is binary rather than exact:

Closing it needs more than exact arithmetic. A decimal division's scale is max(6, s1 + p2 + 1), where p2 is the declared precision of the divisor rather than anything about its value: 1.0 / 3 gives six decimal places and 1.0 / an int column gives twelve, because an int is decimal(10, 0) whatever it holds. Getting that right means carrying declared types through the expression tree, which nothing here does; result_kind says what it is sure of and stops. Half of it, exact for literals and wrong for columns, would be the only approximate thing in the project, so it is left whole and written down instead.

The places themselves are carried where they are known exactly, which is a decimal the query wrote out, a cast to decimal or money, and a variable declared as one. That is what a number is written with when it becomes text, so CONCAT(1.50, '') is 1.50, LEN(1234567.89) is 10, and money is two places whatever it holds: CAST(CAST(12.567 AS money) AS varchar(30)) is 12.57. Each of those used to come out as a float's six significant digits, which made 1.23457e+006 of the second. Nothing infers the places through arithmetic, which is where the divergence above still sits.

Text compares case-insensitively, because every column here is declared SQL_Latin1_General_CP1_CI_AS and a client told one thing and given another has no way to notice. That applies to =, LIKE, IN, DISTINCT, GROUP BY, MIN, MAX, ORDER BY and a join's matching alike.

A join is a hash join on whatever equalities its ON offers, falling back to comparing every pair when it offers none. Two tables of a thousand rows is a million comparisons that way and two thousand with a hash, and these tables arrive from APIs that hand over everything they have. What one would build is capped, so a condition matching everything against everything fails with a message rather than by exhausting memory.

A table can also be written out in the FROM with VALUES, as FROM (VALUES (1, 'a'), (2, 'b')) AS t(a, b). Those values carry no names of their own, so the alias names them, and each column's type comes from what its rows hold. Every bracket after FROM used to be read as a derived SELECT, which this is not, so a client asking for one was told the whole query was unsupported. The three ways of getting the column list wrong answer with SQL Server's own numbers: 8155 where the alias names no columns, 8158 where a row is wider than the list, and 8159 where it is narrower. Either bracketed form can be joined as well as read from, so JOIN (VALUES ...), JOIN (SELECT ...) and the comma form all work; a JOIN used to read a table by name alone and refused both, while both worked in the FROM. Rows of different types settle on one type per column and the rest convert to it, by precedence and not by which row was written first: (VALUES (1), ('2')) reads as two numbers, and (VALUES (1), ('x')) is msg 245 either way round rather than a column of text. A moment outranks text the same way, so text that is not a date beside one is msg 241. What still differs is the type a column is declared as rather than the rows in it: a whole number beside a decimal is numeric(2,1) on a real server and a float here, and a column of only NULLs is int there. Neither comparison can see that, because the batch one reads rows and the query one never writes a constructor.

Named queries, derived tables and subqueries are one mechanism: a SELECT evaluated to a table and then used where a table or a value was expected. A subquery inside a condition is lifted out before the condition is parsed and replaced with a parameter, so the expression layer never learns what a catalog is. Nesting is bounded. A WITH that names itself is recursive, which is not supported: without UNION ALL it is msg 252 as on a real server, and with one it is refused by name.

Integer division truncates toward zero, and division by zero is msg 8134, as it is on a real server.

Table discovery from a client

Not every client finds tables by reading INFORMATION_SCHEMA. Measured against this server:

Without those a client connects, authenticates, and shows an empty table picker, which looks like an empty server rather than a missing procedure. sp_tables, sp_columns and its version-suffixed variants, and sp_databases are implemented in ODBC's documented layout, because drivers read those columns by position as much as by name: one that finds SCALE where it expects RADIX does not report a mismatch, it reports the wrong type. xp_msver, which some clients ask for the server's version, is answered too.

They arrive as RPC calls rather than as SQL text, so which procedures exist is decided by the catalog rather than by the protocol layer. Anything else still answers 2812 rather than completing silently, because a client that asks for a result set and receives nothing raises a NullReferenceException with nothing to explain it.

MSOLEDBSQL asks for a different family. OLE DB defines its own schema rowsets, so sp_tables_rowset2, sp_columns_100_rowset2 and a dozen others exist alongside the ODBC ones with different layouts and a different type system: nvarchar is -9 to ODBC and 130 to OLE DB, and -9 is not a DBTYPE at all. Every layout was read off SQL Server 2025 by running the procedure and taking its result metadata, because a provider reading a column of the wrong type does not report a mismatch.

Talking to an old client

TDS 7.2 widened three fields and added a header block, and a server that ignores the version a client asked for sends the modern shape to everyone. The modern shape is unreadable to an older client: two spare bytes between tokens put every following read at the wrong offset.

The login is answered with the version the client offered, never a newer one. Four other differences from a real server turned up alongside: PRELOGIN answers only the options the client asked about, the SSPI token produced on the completing step is not sent, the session id goes in the packet header from the login response onward, and TLS session tickets are off.

All of that came from one symptom. The legacy "SQL Server" ODBC driver reported a server older than 6.5, then a 7,536,649-byte header block inside a 116-byte packet, then a protocol error in the stream, each error appearing only once the one before it was fixed. Reading a real SQL Server's answer to the same driver through a proxy said what it had done differently each time.

Verified end to end against a catalog of five tables: the legacy ODBC driver, ODBC 17, ODBC 18, .NET SqlClient and MSOLEDBSQL all list 5 tables and 49 columns with types intact, and MSOLEDBSQL answers all 15 of its schema rowsets.

A failed query was the one answer still sent in the modern shape whatever the client asked for. A single error happened to get through the legacy driver, but the two a batch sends for an unclosed quote did not: it reported an unknown token. Both errors now arrive the way a real server sends them to that driver, and the next query on the connection answers.

Protocol notes

Measured against SQL Server 2025 (17.0.1000.7) with Wireshark while desktop Excel connected over Windows Authentication. The captured bytes are the test fixtures, and every token the bridge emits is asserted byte-equal to the one the reference server sent.

Details a client notices and the specification does not make obvious:

• The TLS handshake travels inside packets typed PRELOGIN (0x12). The application records that follow do not; they go on the wire bare.
• Encryption negotiated "off" still encrypts LOGIN7, then reverts to cleartext. The tunnel exists to carry one packet.
• MARS is answered "off", and that answer is honest. A client that asked for multiple active result sets connects and runs; its second open reader is refused by its own driver, which is what happens against a real server with MARS off. Advertising it without the session-multiplex framing that goes with it would be worse than declining, because the client would then wrap every packet in a header this server does not read.
• The TLS context is pinned to 1.2. That is what the reference negotiated, and 1.3 sends NewSessionTicket after the TDS framing has already stopped, where the protocol defines no way to carry it.
• Windows Authentication arrives as SPNEGO wrapping NTLM, offering four mechanisms with NTLM first. The NTLMSSP signature sits 61 bytes into the blob, not at its start.
• The TDS version is stored little-endian in LOGIN7 and big-endian in LOGINACK. Same value, reversed bytes, so echoing back what was parsed is wrong.
• LOGINACK's program name counts 22 characters for "Microsoft SQL Server". The last two are nulls.
• Token lengths are little-endian inside a big-endian packet header.
• The SSPI exchange is asymmetric: the server frames its blob as a token, 0xED plus a little-endian length, while the client sends its blob raw.
• A SQL batch is not just text. It opens with an ALL_HEADERS block that declares its own length, and skipping that by a constant rather than by the declared length puts header bytes into the query string.
• A client's transaction API does not send SQL. BeginTransaction, Commit, Rollback and Save arrive as packet type 0x0E carrying the same ALL_HEADERS block, then a two-byte request type and a name. The reply is an ENVCHANGE holding an eight-byte descriptor the client stamps on every batch it sends inside the transaction. A server that answers that packet with anything else drops the connection, so a .NET client failed at BeginTransaction() before it had sent a query at all.
• The password in a login is not encrypted, it is scrambled: every byte has its nibbles swapped and is then XORed with 0xA5, a constant the specification publishes. Reading one back is the reverse order, XOR and then swap, and doing it in the client's order returns plausible-looking rubbish rather than failing. What keeps a password private on the wire is the tunnel, which covers the login packet even when the connection agreed to encrypt nothing else.
• NULL is spelled differently per type. The one-byte-length types say it with a zero length; nvarchar cannot, because zero is a legitimate empty string, so it spends its whole two-byte length on 0xffff.
• Result columns use the nullable type forms, INTN and FLTN rather than INT4 and FLT8, because only those carry the length prefix a NULL needs.
• Text past 4000 characters cannot declare a size, so it takes the MAX form: the column declares 0xffff and the value arrives as an 8-byte total length, then length-prefixed chunks, then a zero terminator. An API array flattened to JSON reaches this routinely.
• Clients do not send everything as a SQL batch. Anything parameterised, and every catalog query, arrives as an RPC call to sp_executesql with the statement as its first parameter.
• An RPC parameter's value is not self-describing: both its width and its length prefix come from TYPE_INFO, so an unreadable type has to stop parsing rather than be skipped.
• A table picker does not ask for a list of tables. It selects from INFORMATION_SCHEMA.TABLES with a clause like (TABLE_NAME = @Name or (@Name is null)), which needs SQL's three-valued logic to work: with the parameter null the comparison is unknown, not false, and only the IS NULL beside it makes the clause true. Treating unknown as false returns nothing and looks like an empty database.
• The schema part of a name cannot be discarded the way the database part can. INFORMATION_SCHEMA.TABLES and a user table called TABLES are different tables.
• TOP applies after ORDER BY, not before. SELECT TOP 3 ... ORDER BY score DESC means the three highest scores, not three arbitrary rows put in order.
• NULLs sort first ascending and last descending, which is what SQL Server does and not what a naive sort does.
• AVG over an integer column returns a truncated integer. That is what SQL Server does, and it is matched rather than improved so a client computing against both gets the same number. SUM over an integer column is the one deliberate deviation: it widens to 64 bits, because SQL Server's overflow would surface here as an encoding failure partway through a result set rather than as a SQL error.
• An un-aliased aggregate or window function has no column name at all. SQL Server leaves it unnamed and clients render a blank heading, so an empty string is the faithful answer rather than an invented one.
• The PRELOGIN encryption option is a negotiation, not a server setting. A client that asked for ENCRYPT_ON will not read cleartext afterwards, and answering OFF does not fail loudly: it completes the handshake, authenticates, then times out in its post-login phase waiting for bytes that never come.
• No columns means no result set, which is not the same as a query returning no rows. SET and USE produce nothing, and a client sent COLMETADATA for one of those reports an invalid cursor state on the query it was really waiting for.

docs/tds-login-handshake.md has the packet-by-packet breakdown and the implementation order it implies.

Port

Defaults to 1337, not 1433, so it can run beside the SQL Server instance it is checked against.

Data Source=tcp:127.0.0.1,1337

The tcp: prefix is required for a local target. Without it the client selects shared memory and never reaches the socket.

For SSMS, Azure Data Studio and sqlcmd the separator is a comma:

127.0.0.1,1337

A colon is not a syntax error, which is what makes it worth stating. The client reads the whole string as a host name, never sees a port, and fails over to Named Pipes, so the error it reports mentions pipes and a missing network path rather than anything about the port.

Those clients also default to Encrypt=Mandatory and will reject a self-signed certificate, so tick Trust Server Certificate. Encryption itself is fine: the bridge agrees to ENCRYPT_ON and keeps the tunnel up for the whole session.

Development

pip install -e ".[dev]"
python -m pytest

CI installs the same tools from hash-locked files rather than resolving them afresh, and that is the way to get exactly what CI runs:

python -m pip install --require-hashes -r .github/requirements/test.txt
python -m pip install --no-deps --no-build-isolation -e .
python -m pytest

Every pull request runs three checks, and main takes a change only when all three pass: CI (the suite on Ubuntu and Windows for Python 3.10 to 3.14, and the wheel and sdist built and tested), Security and Malware scan. A Fuzz workflow runs the parsers under Atheris, from the seeds in tests/fuzz_corpus, on each change to them and daily. SECURITY.md says what each one checks and how a release is made.

.\scripts\run_dev.ps1            # examples/tables.json on 127.0.0.1:1337
.\scripts\run_dev.ps1 -NoData    # no data source; every query errors

scripts\run_dev.bat is the same thing for cmd or a double-click, which avoids PowerShell's execution policy refusing an unsigned script.

Python 3.10 or newer. The suite needs neither SQL Server nor Excel. The authentication and query tests need Windows, because they run a real SSPI client against a real SSPI acceptor rather than a mock.

scripts/capture_login.ps1 regenerates the reference capture. It needs Wireshark, a reachable SQL Server and desktop Excel, and drives Excel through pyVBAharness. Captures are gitignored: they carry an NTLMv2 challenge and response, which is crackable offline against a weak password.