Skip to main content

epub-blocks

epub-blocks is a dependency-free Python library for reproducibly extracting ordered, structured text blocks from EPUB 2 and EPUB 3 files. A declarative recipe selects an edition’s XHTML, assigns block types, and generates stable identifiers without edition-specific Python code or a pre-existing output table.

It requires Python 3.13 or later. The project is pre-1.0, so its public API and recipe format may still make breaking changes.

Version 0.9.0 adds an optional, experimental terminal wizard for authoring recipes: select files, define references, classify patterns, preview output, and save your decisions. Existing recipes keep their extraction behavior and compiled digests, and recipe format remains "1". Statement and branch test coverage remain at 100%. See the changelog.

Installation

python -m pip install epub-blocks

The core runtime package has no third-party dependencies.

Interactive recipe wizard (experimental)

An optional Textual interface now helps select documents, define section codes, classify block patterns, and preview a draft recipe before saving. It uses no LLM and generates an ordinary version-1 recipe. Unicode inline markup is the default. Install the optional interface with:

python -m pip install "epub-blocks[wizard]"
epub-blocks-wizard book.epub draft.recipe.json

See the wizard guide for save/resume, validation, and the structural cases that still need manual recipe editing.

What it does

The package:

  • safely reads EPUB package metadata and XHTML spine documents;
  • selects source documents and blocks with case-insensitive globs;
  • normalizes extracted text and omits configured EPUB semantic elements;
  • derives groups from source paths, heading markers, or explicit transitions;
  • generates flat and nested identifiers with per-group counters or source paths;
  • assigns output types through ordered source-block rules;
  • handles joins, splits, omissions, skipped blocks, and inserted material;
  • selects whole compound blocks and intentional empty structural blocks;
  • inserts recipe-selected whitespace before/after nested elements without splitting ordinary inline word wrappers;
  • preserves recipe-selected spans and milestones as XML fragments or configurable Unicode delimiters, without requiring new paragraph numbers;
  • pins both the EPUB and the compiled extraction plan with SHA-256; and
  • writes headerless id, type, text TSV records.

A recipe is sufficient to generate a new base text. It does not need a separately stored list of expected identifiers or types. Existing outputs may be useful as test oracles while authoring a recipe, but they are not recipe inputs and are not read by epub-blocks.

The package does not compare editions, select preferred readings, apply editorial corrections, create witness manifests, or emit stand-off annotations. It can generate a marked-up TSV for a separate converter to derive plain text and stand-off together. Recipes describe source structure, not expected prose.

Inspecting an EPUB

Use the lower-level API to discover source blocks before writing a recipe:

from epub_blocks import extract_blocks

for block in extract_blocks(
    "book.epub",
    include_documents=["*chapter*.xhtml"],
    exclude_classes=["image-caption"],
):
    print(block.source_locator, block.tag, sorted(block.classes), block.text)

A locator such as s008:text/chapter-01.xhtml#1.3.2 combines the one-based spine position, EPUB-internal document path, and element path within the XHTML body. Recipes use the EPUB-local portion: text/chapter-01.xhtml#1.3.2.

Recipe example

Recipe version 1 remains the serialized format in epub-blocks 0.9.0. TSV serialization uses literal fields, instead of CSV-style quoting:

{
  "recipe_version": "1",
  "metadata": {"name": "Example edition"},
  "epub": {
    "identifier": "9780000000000",
    "sha256": "f4f9c2d902a41b80732b2dce7ad01a57f615859c21417a5019e3cd8e4d271282"
  },
  "normalization": {
    "collapse_whitespace": true,
    "strip": true,
    "unicode_normalization": "NFC"
  },
  "omit_epub_types": ["noteref", "pagebreak"],
  "source_blocks": {
    "include_documents": ["text/chapter-*.xhtml"],
    "exclude_classes": ["image-caption"]
  },
  "output": {
    "groups": {
      "source_marker": {
        "pattern": "^CHAPTER\\s+([IVXLCDM]+)\\b",
        "case_insensitive": true
      },
      "capture_kind": "roman",
      "capture_width": 2
    },
    "identifiers": {
      "block": {
        "template": "{group}.{number:03d}",
        "start": 1
      },
      "line": {
        "template": "{group}.{block:03d}.{number:02d}",
        "start": 1
      }
    },
    "default": {"type": "paragraph", "role": "block"},
    "rules": [
      {
        "match": {"tag": "h1"},
        "type": "heading",
        "role": "fixed",
        "id": "{group}.000",
        "remove_prefix": {
          "pattern": "^CHAPTER\\s+[IVXLCDM]+\\s+",
          "case_insensitive": true
        }
      },
      {
        "match": {"classes": ["verse-first"]},
        "type": "line",
        "role": "line-start"
      },
      {
        "match": {"classes": ["verse"]},
        "type": "line",
        "role": "line"
      }
    ],
    "skip_source": [],
    "replacements": [],
    "insertions": [],
    "compiled_sha256": "28354978484f35525c616d138bdeb8f801039d665235ac88bd02f1a25791c410"
  }
}

This recipe generates identifiers such as 01.000, 01.001, 01.002.01, and 01.002.02. Rules are matched against source structure; no expected text or identifier table is consulted.

See the complete recipe specification and packaged JSON Schema.

Optional marked-up text

The recipe's text.markup object chooses "format": "xml" or "format": "delimiters". Both serializers use the same structural rules, references, normalized text, and annotation boundaries. For example:

XML:        A <em>quiet</em> word.<page label="iv"/>
Delimiters: A ⧼quiet⧽ word.⟦iv⟧

Milestones between text blocks can attach to the following block without consuming a number. Trailing milestones require an explicit policy. Literal text is escaped so it cannot be mistaken for markup. See the markup contract for the complete configuration, grammar, escaping rules, normalization, slicing, and attachment semantics.

The third format, "literal", supports established notations such as "paragraph": ["¶", "¶"] or "line-break": ["∥", ""] without an escaping layer. It uses the same extraction policy, but source text may collide with markup tokens: literal output is not generally reversible. Use XML or escaped delimiters for unambiguous downstream decoding. Literal mode also leaves raw control characters untouched; the TSV writer still rejects TAB, CR and LF in fields.

Selective whitespace

A recipe can distinguish layout spans from ordinary word wrappers:

"text": {
  "block_boundaries": {
    "tags": ["p", "li"],
    "separator": " ",
    "rules": [
      {"match": {"tag": "span", "classes_all": ["space"]}, "before": " "}
    ]
  }
}

This inserts a space before a retained nested span.space, including an empty one, without discarding its contents. Add after for a trailing space. Other spans remain transparent. Rules are first-match-wins and override the tag shorthand for a matching element. Whitespace enters the character stream before normalization and slicing; optional markup can record its source without losing the space when markup is removed. See the boundary specification and spacing/markup contract.

Command line

epub-blocks book.epub recipe.json records.tsv

The output is headerless TSV with id, type, and extracted text columns. Fields are separated by literal tabs and records by LF. Quotes and backslashes are written unchanged: there is no CSV quoting, quote doubling, or TSV escaping. Embedded TAB, CR, or LF characters in any field are rejected rather than silently altering text or producing ambiguous rows. Normalize text or encode such characters in the markup before writing. A failed write leaves any existing output file intact.

Python API

from epub_blocks import extract_recipe_file

records = extract_recipe_file("book.epub", "recipe.json")
for record in records:
    print(record.block_id, record.block_type, record.text)

compile_recipe_file returns the immutable in-memory extraction plan. For recipe authoring, extract_recipe_candidates(epub_path, recipe) returns source blocks with locators, tags, classes, and plain normalized text, using the recipe's source-element and text policies. It checks the EPUB hash and identifier policy but does not apply output rules or require/verify the compiled-plan digest. A syntactically valid draft output section is still required; use load_recipe to read JSON. Final output references, coverage, and detached milestone placement are checked by compilation, not candidate inspection. Since 0.6.0, a recipe may omit epub.identifier only for an EPUB with no nonempty package identifiers; the archive SHA-256 is still mandatory. A supplied mismatched identifier never falls back to hash-only checking. The compiled plan's epub_identifier is None in this case. See the source-pin contract. Fragment and extract_fragments provide lower-level access to selected XHTML subtrees. The supported import surface is the names exported by epub_blocks.

Safety

EPUB files are untrusted ZIP and XML input. Default APIs bound archive size, individual and cumulative reads, compression ratios, XML size, element count, and nesting depth. Duplicate or unsafe paths and encrypted members are rejected. External DTDs are never loaded; nonempty internal subsets, entity declarations, other doctypes, and unknown named character references are rejected. Version 0.7.0 additionally accepts an empty internal subset ([]), including XML whitespace between the brackets. The declaration is stripped, not interpreted; no custom entities or DTD rules are enabled.

Recipes are trusted local configuration because they contain regular expressions. epub-blocks never executes EPUB scripts or fetches resources from the network.

Development

uv sync
uv run ruff format --check .
uv run ruff check .
uv run pyright
uv run coverage erase
uv run coverage run -m unittest discover -s tests
uv run coverage report
uv run python -m build
uv run twine check --strict dist/*
uv run pyright --verifytypes epub_blocks --ignoreexternal

Coverage includes branches and enforces a 100% minimum. Tests build synthetic EPUBs; no EPUB files are committed. See CONTRIBUTING.md and CHANGELOG.md.

License

epub-blocks is available under the MIT License.

Metadata

Release files for epub-blocks 0.9.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for epub-blocks 0.9.0
File Size Uploaded Provenance Actions
epub_blocks-0.9.0.tar.gz 188.2 kB With provenance Details for epub_blocks-0.9.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for epub-blocks 0.9.0
File Interpreter ABI Platform Actions
epub_blocks-0.9.0-py3-none-any.whl Python 3 none any Details for epub_blocks-0.9.0-py3-none-any.whl

Total release size: 260.7 kB

Release files / epub_blocks-0.9.0.tar.gz

Download URL epub_blocks-0.9.0.tar.gz
Size 188.2 kB
Tags Source
SHA-256 checksum
How to use checksums
27961776ce430552da4a805b510954dfad1a996e36090f9fea78fa92b4372893
BLAKE2b-256 checksum
How to use checksums
ba24671ebf383038f9d54d48be57ce3d814f2bfefb9db4ca0e717a86eb1c7738
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 1, 2026.

Transparency log

Release files / epub_blocks-0.9.0-py3-none-any.whl

Download URL epub_blocks-0.9.0-py3-none-any.whl
Size 72.5 kB
Tags Python 3
SHA-256 checksum
How to use checksums
8796e8386b5b15340353694ef92167982d2a5fa06b474ee9a9c5502e00ff30cb
BLAKE2b-256 checksum
How to use checksums
b018451fabff3511c2383e0069fd4aa5f7730816876be200c04eb158fbfee2be
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 1, 2026.

Transparency log

Security for epub-blocks 0.9.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for epub_blocks-0.9.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 1, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page