mock-acme

The integration between the mocks. mock-sap, mock-edi, mock-bank and mock-einvoice stand in for systems you do not own, so that the code between them can be tested. This is that code: ACME's middleware, the fourth actor in every story the mocks tell.

It is not a mock, and it is not a client library to put in front of real money. It is a reference: worked integrations, each with tests that run against the released mocks, kept in one place so that there is one copy of each.

What is here

mockacme.bank_messages is what the two payment modules share: the call to the bank and the reading of its ISO 20022 answers.

Every module talks to the mocks over HTTP and imports none of them. The package has no dependencies.

Installing

python3 -m pip install mock-acme

That installs the package and nothing else. The mocks it talks to are separate: pip install mock-sap mock-edi mock-bank mock-einvoice.

payment_run needs mock-sap 0.19.0 or later. It writes which payment run has an invoice on the invoice itself, which an older mock-sap refuses, so nothing is paid. And against an older one, a customer's payment that quotes the number of an invoice already paid reopens that invoice.

Tested against

What each release was tested against: the mocks CI installed for the run on the release's own commit.

A row is what was run, and nothing else. It does not say an earlier or a later mock fails. The lowest versions the tests are known to need are the floors in the test extra in pyproject.toml, and nothing runs the tests at those floors, so they say "not below this" and no more.

Rows are not typed in. tools/tested_against.py writes one from the mocks installed where the tests have just run, and publish.yml refuses to publish a release whose row is not what its own run installed. The rows up to 0.5.0 are older than the tool, and were read from the CI logs of each release commit.

Running the tests

From a checkout:

python3 -m pip install -e ".[test]"
python3 -m unittest discover -s tests -t . -v

That is the whole arrangement. Importing tests starts the four mocks, each on a port the operating system chose, and stops them afterwards. tests/__init__.py says how to point the tests at a mock that is already running, which is how they are run against a mock's main.

tests/test_payment_run_readers.py is the one test that imports a mock rather than talking to it: it gives payment_run's hand-written statement readers the same statement mock-bank wrote as a camt.053 and as BAI2, and five BAI2 files from moov-io/bai2, and holds them to mock-bank's own reader. The samples and their licence are in tests/samples/external/.

payment_run pays by ACH as well. The same tests in that mode:

PAYMENT_RUN_FORMAT=nacha python3 -m unittest -v tests.test_payment_run

The released mocks are what a pull request is tested against. A second workflow, mocks-main.yml, runs the same tests every night against each mock's main, so that a change merged there which breaks this code is seen before it is released. It is not a required check; a red run there says something about another repository. To do the same by hand:

python3 -m pip install \
  "mock-sap @ git+https://github.com/rseufert/mock-sap@main" \
  "mock-edi @ git+https://github.com/rseufert/mock-edi@main" \
  "mock-bank @ git+https://github.com/rseufert/mock-bank@main" \
  "mock-einvoice @ git+https://github.com/rseufert/mock-einvoice@main"

Known to be wrong

Two payment runs that select at the same moment can both pay an invoice. A run says in SAP which run has each item before its file goes out, as PaymentRunID and PaymentRunDate on the invoice, and every other run leaves that item alone until the bank refuses the payment, SAP clears it, or the payment comes back (#2, #21). But reading the open items and writing the claim are two requests, and nothing makes them one: two runs that both read before either writes both pay. One run after another, on any machine, is safe; two at once are not.

An e-invoice posted by a process that stopped before saying so is never answered. e_invoice keeps what it has collected and said in memory, and a new process asks the supplier's side what each invoice was last told, so nothing is acknowledged, accepted or rejected twice, and an invoice accepted before a restart is still told "paid". But when the last thing said was "received" and SAP already holds the invoice, there are two ways that came about: the earlier process posted it and stopped before it said "accepted", or the invoice is a copy of one that reached SAP another way. SAP holds a number and a supplier, not who posted it. So the invoice is reported as blocked with both reasons, the supplier is told nothing, and "paid" is never said for it. A person has to look. An XRechnung invoice, which is told nothing, is reported as blocked for being in SAP after every restart, for the same reason.

What a shipment has been billed is one process's memory. invoice_check counts each invoice against the shipment it names, so a supplier that bills one delivery twice under two invoice numbers has the second blocked (#41). A restart loses the count, in DurableInvoiceCheck as well: it asks SAP what an order has been billed, and SAP's supplier invoice has nowhere to say which shipment it was for, in mock-sap and so here. A new process holds such an invoice for a ship notice it never saw; if the supplier sends that notice again, the invoice is posted. And an invoice that names no shipment is not counted against one at all: two of them for parts of one delivery are each compared with the order's latest notice on its own. mock-edi sends neither a second invoice for a shipment nor one that names none, so the tests write them.

A held invoice waits for ever if its ship notice never comes, and is reported as held on every run. What is held, like what has shipped, is one process's memory: a restart loses it, and the invoice with it, since collecting from the mailbox took it out.

It has other edges:

• An e-invoice names no shipment, so e_invoice never holds one: an e-invoice for more than the order's latest ship notice is rejected.

• e_invoice does not read allowances or charges, so an e-invoice with one is rejected because it does not add up. It does not read a credit note at all. A part payment is not told as one: a payable is cleared or it is not.

• Which of Peppol's reason codes an e-invoice is rejected with is this package's reading: its list has no code for a duplicate, which is told as a wrong reference (REF), nor for a currency, which is told as other (OTH).

• A run's identification is one to six characters, which is SAP's own limit. A longer one is refused before anything is selected.

• When the bank refuses a payment and SAP cannot then be reached, the claim stays on the item and no run selects it. The run says so as a problem, and a person has to take the claim off.

• Register, the run's own record from before SAP could hold this, is still accepted and no longer needed. Passing one means two records of one fact: whoever posts the statements has to be given the same register, or its entry outlives the claim and holds an item SAP says is free.

A payment the bank reports without its reference is not cleared. A statement line that quotes nothing goes to SAP quoting nothing, mock-sap matches no item to it (it does not fall back to the amount), and the invoice stays open and claimed by the run that paid it, so no later run pays it either (#44). The run says so as a problem, with SAP's reason and its own accepted payments of that amount (#46), and a person has to clear the item or take the claim off. It does the same for any debit SAP places nowhere, which includes one that was never this run's: the run cannot tell. mock-bank always gives the reference back, so the first case takes a statement from somewhere else.

Money arriving clears the receivable it quotes, where SAP can tell which that is, and where it cannot the money is applied to nothing and order_to_cash says so with SAP's reason (#42). What that leaves:

• A customer who pays short, or over, is applied to nothing. mock-sap posts no part payment and no residual item, and nothing here decides whether a shortfall is a deduction or an instalment.
• A payer who writes the invoice number in the EndToEndId and something else in the remittance information is not applied. The remittance information is believed over the EndToEndId, which is the payer's own number for the payment; only a payer who writes no remittance information at all has the EndToEndId read.
• A BAI2 statement carries no note to payee, so on an account that banks in NACHA only a structured reference or the EndToEndId can name the invoice.
• order_to_cash writes an invoice at one rate of tax, in the standard category, for one sales order. A billing document with no tax, or over several orders, is not sent, and neither is a credit memo or a cancellation. Nobody is reminded when an invoice is past due.
• Who is selling, the rate of tax and where each customer takes its invoices are given to order_to_cash and not read from SAP: mock-sap has no company code master, no rate on a billing document and no e-invoicing address on a business partner. A real system holds all three somewhere.

Releasing

A release is the same three acts as in the mocks, by one person in one sitting: merge a pull request that sets version in pyproject.toml and __version__ in mockacme/__init__.py and dates the section in CHANGELOG.md, and has the release's row under Tested against, written by python3 tools/tested_against.py --write after the tests have passed; tag that commit v<version>; publish a GitHub Release from the tag. Publishing the Release runs publish.yml, which runs the tests, builds, refuses a tag that disagrees with the package or a "Tested against" row that disagrees with the mocks it just tested with, and uploads to PyPI through Trusted Publishing. Running that workflow by hand uploads to TestPyPI instead.

There is no tools/release.py here as there is in the mocks: the steps are done by hand. CI checks that the two places the version is written agree, and that the version has a row under "Tested against".

Where the code came from

The modules were copied on 2026-10-04 from the examples/ folder of the mock each was written beside, with their tests; remittance was written there afterwards and followed on 2026-10-05. Nothing in them changed but imports, the lines saying how to run the tests, and issue references, which now name the repository they belong to.

The originals were removed from those repositories on 2026-10-05, and each left an examples/README.md saying which file became which. This is the only copy.

The repository's settings are described in docs/GITHUB.md.