Raven-1/tests/test_knowledge_structure.py

438 lines
16 KiB
Python
Raw Permalink Normal View History

fix(tools): an array sent as its own JSON is still an array ## Summary Reported as "deliver_files is broken". It is not: **the tool never ran.** Every `deliver_files` call on one install was refused before dispatch -- five in a row in a single session -- and nothing was ever delivered: ``` Error: Invalid parameters for tool 'deliver_files': files should be array ``` What the model sent was the array, spelled as a string: ```json {"files": "[{\"path\": \"...poster.png\", \"title\": \"...\"}]", "message": "..."} ``` `cast_params` already forgives that spelling for every scalar type -- `"3"` for an integer, `"true"` for a boolean -- and did not for arrays or objects, which are exactly the types a model is most likely to stringify. ## Type - [x] Fix - [ ] Feature ## The edge of the forgiving is the design When the value is a string, try `json.loads`, and take the result **only if it is the type the schema asked for**. A string that is not JSON, or is JSON of the wrong shape, is returned unchanged so the validator still names the real problem -- the point is to widen what is accepted, not to swallow a type error. **Nothing here can raise.** Turning "your argument has the wrong type" into a traceback would be worse than the refusal it replaces, so the parse is wrapped and failure falls through. **A `type: "string"` field is never parsed.** JSON is ordinary text in a message body or a file's content, and replacing it would hand the tool something it never asked for. Two independent things prevent it today (the string branch returns before the parse is reached, and the parse only wins on a type match), which is why the case pinning it is a tripwire rather than a load-bearing assertion -- said so in its own docstring rather than left to look like coverage. ## Not just deliver_files Six built-in tools declare an array or object parameter and were all exposed: | tool | parameter | |---|---| | `deliver_files` | `files` | | `ask_user` | `questions` | | `image_generate` | `images` | | `message` | `media` | | `create_playbook` | `skills` | | `video_generate` | `params` | The same failure was already seen from the other side: `acp_client/acp_agent.py` records `update_task_state` and `operations should be array` from a real dispatch. ## Verification ``` make lint-python ruff check + ruff format --check, 1850 files make lint-imports 9 contracts kept, 0 broken pytest tests/ 19,699 passed, 7 failed (all pre-existing, see below) ``` - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [ ] User-facing docs or screenshots -- no user-facing string **End to end, through the real tool**, with the exact payload that failed five times in production: ``` files sent as: str -> [{"path": ".../poster.png", "title": "..."}] returned: Delivered 1 file(s): poster.png (72 B). still "should be array": False registered in the store: True ``` Twelve cases, five mutations, each killing at least one: | mutation | red | |---|---| | drop the whole branch (the old behaviour) | 3 | | raise instead of falling through on a parse failure | 5 | | accept the parse whatever type it yielded | 1 | | do not re-cast the parsed value, so items are not cast | 1 | | widen the tuple to include `string` | **survived -- unreachable** | The last is worth stating rather than hiding: that mutation cannot change behaviour, because the `type == "string"` branch returns earlier and the type check would refuse the result anyway. I also tried moving the new block above that branch, alone and combined with widening the tuple, and neither reddens anything either. The invariant holds by construction; the case that documents it says so. **Pre-existing failures, checked against the base.** 7 failed in the full run, all in `tests/test_security_network.py` (`test_a_redirect_to_a_dangerous_address_is_refused` and siblings). Running that file alone on this same tree gives 4 -- the count is order-dependent -- and the change touches only `raven/agent/tools/params.py` plus a new test file, neither of which `raven/security/network.py` imports. The failures are the box's resolver: it returns poisoned answers roughly half the time, which I measured separately (`en.wikipedia.org` came back as `2001::1` on 3 of 6 lookups). Not introduced here, and worth someone's attention on its own. ## Risk - [x] Security impact considered -- widens what reaches a tool's own validation, not what bypasses it. A parsed value is still validated against the same schema, and a parse that yields the wrong type is refused exactly as before - [x] Backward compatibility considered -- a call that worked before takes the identical path; the new branch is reached only by a `str` where an array or object is declared, which previously always failed - [x] Rollback path is clear -- single-commit revert ## Related Issues N/A ## Swept `mr-review-patterns` pre-submit sweep, over `git diff origin/main...HEAD`. - **1 (a write location moved)** -- nothing moved; a branch was added to an existing dispatch on `target_type`. Grepped every caller of `cast_params`: the registry is the only one, and it runs `validate_params` right after, which is the reader this affects. - **2 (a value with several entry points)** -- the cast is the single funnel every tool argument passes through, which is why the fix belongs here and not in `deliver_files`. - **3 (a guard)** -- checked both directions: what it wrongly lets through (JSON of the wrong type, non-JSON text -- both refused, with cases) and what it wrongly refuses (a real list, unaffected, with a case). - **4 (tests)** -- five mutations above, one honestly reported as unreachable. Failing-ID sets diffed against the base rather than counts. - **6 (stale text)** -- the module docstring enumerated what is coerced ("a string that should be an integer"); it now names arrays and objects too. Grepped the tracked tree for other statements of the rule: none. - **9 (error handling)** -- the new path adds one way to end and it is caught: `json.loads` raises `ValueError` on bad text and `TypeError` on a non-string, and both fall through to the untouched value. - **11** -- `git ls-files | grep -c node_modules` is 0; the commit-message gate was run, not eyeballed. Not applicable, with reasons: **3b, 3c, 3d, 3e** (no tool schema produced, no state moved, one lane, no assertion lost its symbol); **4b, 4c, 4d** (no platform branch, no shell, no rendered number); **5, 5b** (no dependency; `json` is stdlib and already imported across this package); **6b, 6c** (no backup, no generated artifact); **7, 7b** (nothing on disk); **8** (no asset, no secret); **10** (no prompt text). ## Deliberately not fixed The error text still ends with `[Analyze the error above and try a different approach.]`, and the model plainly did not -- it sent the same shape five times. A message naming the expected spelling (`files must be a JSON array, not a string containing one`) would give it something to act on. That is a change to the registry's error surface, affects every tool, and deserves its own argument. ## Moved from !594 Opened on a compliant source branch. AGENTS.md 2.1 asks for a 3-5 word snake_case description and the previous branch had seven; GitLab cannot change an existing merge request's source branch, so the change moves rather than the name. Same commits, rebased onto `c2a9768f`. Both code findings from the review on !594 are already in them -- `RecursionError` caught with its regression case, and the implementation comment reduced to the two non-obvious reasons. Nothing else changed, so that review carries.
2026-09-09 15:42:36 +00:00
"""Unit tests for heading-aware sectioning and chunking."""
from __future__ import annotations
import asyncio
from raven.knowledge._sections import MAX_SECTION_CHARS, SECTION_ORDINAL
from raven.knowledge._structure import HeadingAwareChunker, StructuredTextParser
from raven.knowledge._types import Section, TextBlock
def _parse(text: str, filename: str) -> list[Section]:
return asyncio.run(StructuredTextParser().parse(text, filename))
def _chunk(sections: list[Section], **kwargs) -> list:
return asyncio.run(HeadingAwareChunker(**kwargs).chunk(sections))
def _texts(items) -> list[str]:
return [item.content.text for item in items]
MARKDOWN = """\
Intro paragraph.
# Results
Overall summary.
## Latency
The p99 was 40ms.
## Throughput
Peaked at 900 rps.
# Appendix
Raw numbers.
"""
def test_markdown_splits_on_headings():
sections = _parse(MARKDOWN, "report.md")
assert [section.metadata.get("heading") for section in sections] == [
None,
"Results",
"Latency",
"Throughput",
"Appendix",
]
assert "Intro paragraph." in sections[0].content.text
assert sections[0].metadata == {}
def test_markdown_records_the_ancestor_path():
sections = _parse(MARKDOWN, "report.md")
latency = next(s for s in sections if s.metadata.get("heading") == "Latency")
assert latency.metadata["heading_path"] == ["Results", "Latency"]
assert latency.metadata["heading_level"] == 2
def test_a_sibling_heading_pops_the_deeper_level():
"""`Appendix` is an h1 arriving after two h2s; leaving them on the stack
would file the rest of the document under a subsection it left."""
sections = _parse(MARKDOWN, "report.md")
appendix = next(s for s in sections if s.metadata.get("heading") == "Appendix")
assert appendix.metadata["heading_path"] == ["Appendix"]
def test_markdown_keeps_its_heading_markers_in_the_section_text():
sections = _parse(MARKDOWN, "report.md")
latency = next(s for s in sections if s.metadata.get("heading") == "Latency")
assert latency.content.text.startswith("## Latency")
def test_a_hash_inside_a_code_fence_is_not_a_heading():
text = "# Real\n\n```bash\n# not a heading\necho hi\n```\n"
sections = _parse(text, "snippet.md")
assert len(sections) == 1
assert "# not a heading" in sections[0].content.text
def test_a_hash_without_a_space_is_not_a_heading():
sections = _parse("#hashtag stays inline\n", "note.md")
assert len(sections) == 1
assert sections[0].metadata == {}
def test_a_heading_holding_only_subsections_is_not_indexed_alone():
"""It would be a chunk of nothing but its own title; the children keep the
name in their path."""
sections = _parse("# Parent\n## Child\n\nBody.\n", "doc.md")
assert [s.metadata.get("heading") for s in sections] == ["Child"]
assert sections[0].metadata["heading_path"] == ["Parent", "Child"]
def test_a_document_without_headings_stays_one_section():
sections = _parse("Just prose.\nMore prose.\n", "plain.md")
assert len(sections) == 1
assert sections[0].metadata == {}
assert sections[0].source == "plain.md"
def test_an_empty_document_still_yields_one_section():
sections = _parse(" \n", "empty.md")
assert len(sections) == 1
HTML = """\
<html><head><title>T</title><style>h1 { color: red }</style></head>
<body>
<h1>Findings</h1>
<p>Overall&nbsp;good.</p>
<h2>Errors</h2>
<p>Three of them.</p>
<script>console.log("noise")</script>
<table><tr><td>a</td><td>b</td></tr></table>
</body></html>
"""
def test_html_splits_on_heading_tags():
sections = _parse(HTML, "report.html")
assert [s.metadata["heading"] for s in sections] == ["Findings", "Errors"]
assert sections[1].metadata["heading_path"] == ["Findings", "Errors"]
def test_html_tags_and_scripts_do_not_reach_the_index():
sections = _parse(HTML, "report.html")
body = "\n".join(_texts(sections))
assert "<p>" not in body
assert "console.log" not in body
assert "color: red" not in body
assert "Overall good." in body
def test_html_is_detected_without_an_html_extension():
sections = _parse("<h1>Title</h1><p>Body.</p>", "export")
assert sections[0].metadata["heading"] == "Title"
def test_a_markdown_extension_is_never_sniffed_as_html():
"""A fenced HTML example inside a .md file must not switch parsers."""
sections = _parse("# Doc\n\n```html\n<html><h1>x</h1></html>\n```\n", "guide.md")
assert sections[0].metadata["heading"] == "Doc"
assert "<html>" in sections[0].content.text
def test_chunks_never_span_two_sections():
chunks = _chunk(_parse(MARKDOWN, "report.md"))
assert all("Latency" not in text or "Throughput" not in text for text in _texts(chunks))
assert [chunk.chunk_index for chunk in chunks] == list(range(len(chunks)))
assert {chunk.total_chunks for chunk in chunks} == {len(chunks)}
def test_a_split_section_carries_its_heading_path_on_later_chunks():
long_body = "word " * 400
sections = _parse(f"# Top\n\n## Deep\n\n{long_body}\n", "long.md")
chunks = _chunk(sections, chunk_size=64, overlap=8)
assert len(chunks) > 1
assert chunks[0].content.text.startswith("## Deep")
assert all(chunk.content.text.startswith("Top > Deep\n\n") for chunk in chunks[1:])
def test_a_headingless_section_is_chunked_unchanged():
sections = [Section(content=TextBlock(text="a " * 400), source="plain.txt")]
chunks = _chunk(sections, chunk_size=64, overlap=8)
assert len(chunks) > 1
assert not any(chunk.content.text.startswith(" > ") for chunk in chunks)
assert all(chunk.metadata == {} for chunk in chunks)
def test_the_prefix_is_paid_for_out_of_the_chunk_budget():
"""Prefixing after the split would push a chunk past the size the embedding
model was configured for."""
sections = _parse("# A\n\n## B\n\n" + "word " * 400, "long.md")
chunks = _chunk(sections, chunk_size=64, overlap=8)
assert all(len(chunk.content.text.encode("utf-8")) // 4 <= 64 for chunk in chunks)
def test_a_split_section_stores_a_verbatim_copy_of_itself():
"""Reassembling a section from its chunks after the fact is unsafe on
repetitive text, so the split records the original instead."""
body = "One clause covering the cell, the system and the harness. " * 40
section = _parse(f"## Deep\n\n{body}\n", "long.md")[0]
chunks = _chunk([section], chunk_size=64, overlap=8)
stored = [c.metadata.get("section_text") for c in chunks if "section_text" in c.metadata]
assert len(chunks) > 1
assert stored == [section.content.text]
def test_a_section_that_fits_in_one_chunk_stores_no_copy():
"""The chunk already is the section; a copy would double it for nothing."""
chunks = _chunk(_parse("## Small\n\nShort body.\n", "s.md"))
assert all("section_text" not in chunk.metadata for chunk in chunks)
def test_an_oversized_section_is_not_copied():
"""It would never be inlined, so storing it only inflates the index."""
section = _parse("## Big\n\n" + "x" * (MAX_SECTION_CHARS + 100), "big.md")[0]
chunks = _chunk([section], chunk_size=64, overlap=8)
assert all("section_text" not in chunk.metadata for chunk in chunks)
# ------------------------------------------------------------ section identity
def test_each_section_gets_its_own_ordinal():
sections = _parse("# A\n\nbody a\n\n# B\n\nbody b\n", "doc.md")
assert [s.metadata[SECTION_ORDINAL] for s in sections] == [0, 1]
def test_two_same_named_siblings_are_told_apart():
"""The failure no guard on the reading side could close: one parent with two
same-named children gives both the same full heading path, and a reader keying
on the path hands a hit in the second the text of the first. The ordinal is
what separates them, and it can only be recorded here."""
text = "## Usage\n\nlead\n\n### Example\n\nfirst body\n\n### Example\n\nsecond body\n"
sections = _parse(text, "doc.md")
examples = [s for s in sections if s.metadata.get("heading") == "Example"]
assert len(examples) == 2
assert examples[0].metadata["heading_path"] == examples[1].metadata["heading_path"]
assert examples[0].metadata[SECTION_ORDINAL] != examples[1].metadata[SECTION_ORDINAL], (
"the two sections share an identity"
)
def test_a_document_with_no_headings_records_no_ordinal():
"""One section cannot be ambiguous, and the no-headings case is documented to
degrade to the previous behaviour exactly -- empty metadata included."""
sections = _parse("just prose, no headings at all\n", "doc.md")
assert len(sections) == 1
assert sections[0].metadata == {}
def test_every_chunk_of_a_section_carries_its_ordinal():
"""The ordinal has to reach the chunks, not only the sections: expansion reads
it off a chunk to find the section's siblings."""
long_body = "sentence. " * 400
text = f"## Usage\n\nlead\n\n### Example\n\n{long_body}\n"
chunks = _chunk(_parse(text, "doc.md"))
example = [c for c in chunks if c.metadata.get("heading") == "Example"]
assert len(example) > 1, "this case needs a split section"
assert {c.metadata[SECTION_ORDINAL] for c in example} == {1}
def test_the_structured_parser_only_claims_what_it_can_cut():
"""This parser claims only the two formats it can find headings in, so a
registry holding it alone would leave plain text, CSV, JSON, YAML and RST
with no parser at all. TextParser has to stay alongside it for the rest."""
feat(knowledge): read every format, show what the index holds, and let an agent search it (#489) <img width="2318" height="1448" alt="image" src="https://github.com/user-attachments/assets/bca71d69-4972-420c-b14e-c8cf74a17de9" /> ## Summary Everything a knowledge base could not read, and the pieces it cut them into saying what they are. **Reading a PDF.** A PDF records glyphs at coordinates; whether they are a heading, a table cell or a running footer is inferred from how they look. There are now two readers. The geometric one infers structure from type size and position and needs nothing but the file. The other asks deepdoc's models, ported into `raven/knowledge/parser/deepdoc`: a layout model that says what each region is, a detector and recognizer that read a page carrying no text at all, and a table-structure model that rebuilds a grid from a picture of one. Detecting boxes around text the file already spells out exactly costs an inference pass per page to learn less, and recognising it costs accuracy. The layout graph answers with a shape its own postprocessing raises on, and its classes are DocLayout-YOLO's order rather than the list Both were established by running the model, and the label order is pinned by a test. **Tables go in as tables.** `<table>` is the only shape that keeps what a grid states -- which column a cell stood under, which cells are merged, which row is the header -- and it is the shape a model reads a table in, which is what these chunks are for. PDF, Word and PowerPoint all state their merges exactly, and they now survive as `colspan` and `rowspan`. The flat form this replaces had to repeat a header standing over two columns across both of them, because there was nowhere to record a span; it had no callers left afterwards and is gone. Two deviations from upstream, both about what the string is for: every cell's text is escaped, where upstream interpolates it raw and a cell holding a `<` produces markup that is no longer a table, and the span attributes are written only where there is a span, where upstream emits `<td >` for every ordinary cell. **A picture of where each chunk came from.** A chunk is a few hundred words with a page number on it, which is enough to cite and not enough to answer whether the parser read the page correctly. So every PDF chunk that knows where it sits carries one image: the regions it covers, cropped out of the rendered pages and stacked. here they are already structured in the parser's element spans, so the markers would only put text nobody wrote into the thing being embedded. The panel draws a chunk's tables as tables, parsed and rebuilt as elements rather than injected: a person can rewrite any piece from that same panel, so what comes back from the store is user input however it got there. **The page that reads them.** Upstream retired the knowledge page while this branch was out; it is rebuilt against the new architecture, and the half that never existed is the one a reader needs -- what a document actually became once it was cut and embedded. Documents sit in one-level folders and arrive by four routes: a file, a folder, a URL, a written note. Adding and indexing are two calls on the contract and nothing drains that queue on its own, so a page that stopped after the first left every upload saying Queued for good. Opening a file takes the whole page: the document on the left, the pieces the index holds of it on the right. The left half is the file drawn as its own pages rather than framed by the browser's PDF viewer, which reads where to open once and answers to nothing afterwards -- and which nothing can be drawn on. Pages are rendered once and kept, and the region a piece was cut from is washed over the page it came from, placed in fractions of the page so the picture's own scale never enters the arithmetic. Clicking a piece takes the file there; double-clicking opens it for rewriting. The pieces can be searched, filtered by state, turned on and off, deleted, rewritten and added to, one at a time or in a batch. Every one of those was already on the contract with no reader for it. A base's settings are back and fuller: how many pieces a search returns, how documents are cut, and how much prose comes with a table or a figure. The embedding model is a picker while the base holds no pieces and a stated fact once it does -- moving it then is the collection made again at the new width and every document re-cut, which is not a thing a panel does by being saved. **An agent can read them.** The retrieval engine was there and nothing could reach it: no agent had a tool that opened a library. A reader now points a conversation at bases from the composer, the way they pin a working directory, and the selection rides in the same place -- the session's metadata, bound around the turn. The tool takes a query and never a base. The scope is the reader's choice, so a model cannot reach material nobody offered it, and it is withheld where nothing was picked rather than offered and refused. A hit says which document and page it came from, because a hit a reader cannot place is a hit they cannot check -- and the passages are pressable in the transcript, walking into the base, the document, the folder it is filed under and then the piece. A citation a reader cannot follow is a claim they have to take on trust, and the whole point of searching their own material is that they can go and look. **The word segmenter and the resources.** `raven.core.Tokenizer` is Infinity's `rag_tokenizer`, ported: a forward and backward maximum match over a dictionary trie, with stemming and lemmatisation on the latin path. XGBoost is not carried over. None of the data lives in git -- the dictionary, the NLTK corpora and the ONNX graphs run to about a hundred megabytes, so `make fetch-resources` on an install and an optional Docker layer behind `RAVEN_RESOURCES` both point at one script. A checkout that could not reach the network still works, and every feature that reads this data says so when asked. ## Type - [ ] Fix - [x] Feature - [ ] Docs - [ ] CI / tooling - [ ] Refactor - [ ] Other ## Verification Run on macOS against Python 3.12, with the deepdoc weights fetched (the model tests skip without them): ``` uv run pytest 27450 passed, 111 skipped, 3 failed npm run test --prefix ui-web 3276 passed npm run test --prefix ui-tui 2106 passed npm run type-check --prefix ui-web clean (tsc) npm run lint --prefix ui-web clean (4 pre-existing hook warnings) make lint-python clean (ruff check + format) make lint-types clean (ty) make lint-imports 10 contracts kept, 0 broken make check-source-language clean make check-large-files clean make build-ui built, both boot snapshots match node ui-web/scripts/check-css.mjs clean node ui-web/scripts/check-class-namespace.mjs clean ``` The three Python failures are not this branch's and were each confirmed on the base alone: one asks a host without LibreOffice to refuse a render and this host has LibreOffice installed, and two simulation cases fail on an unmodified checkout of the base commit. The RPC contract was regenerated for both consumers and both match it (`npm run gen --prefix ui-web`, `npm run gen:rpc --prefix ui-tui`, then `gen:check`). The page's boot snapshots and region goldens were refreshed where the composer bar gained a button, and each records that button and nothing else. - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [x] User-facing docs or screenshots are updated when needed ## Risk A PDF's tables, and a Word or PowerPoint file's, are now indexed as HTML rather than as header-prefixed lines. Chunk text changes shape for those documents, so a base indexed before this and searched after it holds both forms until its documents are reindexed; retrieval keeps working either way, because both forms are text. The same applies to how much prose a table or a figure carries with it: the setting now reaches as far as it says, where before it stopped at the nearest paragraph, so a chunk cut before this keeps the narrower context until its document is rebuilt. Rolling back is reverting the branch and reindexing. The deepdoc models are optional. An install that never ran `make fetch-resources` reads PDFs geometrically, exactly as before, and says so when the models are asked for by name. Nothing here reaches the network at request time. Chunk crops and rendered pages are written at index time and on first view, both under the state directory rather than in the repository, and both are deleted with their document. A page is immutable for the life of its document id, so the route serving them caches; the document route does not. All three routes take ids and no path, and the two image routes answer under the strictest sandbox in the tree. What an agent may search is the reader's choice and nothing else. The tool takes no base argument, so a model cannot name one, and it is absent from the schema entirely for a conversation that named none -- a lane nobody pointed at a base is unchanged by this branch. The bases a session carries are stored beside its working directory and travel no further. - [x] Security impact considered - [x] Backward compatibility considered - [x] Rollback path is clear for risky changes ## Related Issues N/A --------- Co-authored-by: Claude (claude-opus-5) <noreply@anthropic.com>
2026-10-12 19:05:16 +08:00
from raven.knowledge.parser.text_parser import TextParser
fix(tools): an array sent as its own JSON is still an array ## Summary Reported as "deliver_files is broken". It is not: **the tool never ran.** Every `deliver_files` call on one install was refused before dispatch -- five in a row in a single session -- and nothing was ever delivered: ``` Error: Invalid parameters for tool 'deliver_files': files should be array ``` What the model sent was the array, spelled as a string: ```json {"files": "[{\"path\": \"...poster.png\", \"title\": \"...\"}]", "message": "..."} ``` `cast_params` already forgives that spelling for every scalar type -- `"3"` for an integer, `"true"` for a boolean -- and did not for arrays or objects, which are exactly the types a model is most likely to stringify. ## Type - [x] Fix - [ ] Feature ## The edge of the forgiving is the design When the value is a string, try `json.loads`, and take the result **only if it is the type the schema asked for**. A string that is not JSON, or is JSON of the wrong shape, is returned unchanged so the validator still names the real problem -- the point is to widen what is accepted, not to swallow a type error. **Nothing here can raise.** Turning "your argument has the wrong type" into a traceback would be worse than the refusal it replaces, so the parse is wrapped and failure falls through. **A `type: "string"` field is never parsed.** JSON is ordinary text in a message body or a file's content, and replacing it would hand the tool something it never asked for. Two independent things prevent it today (the string branch returns before the parse is reached, and the parse only wins on a type match), which is why the case pinning it is a tripwire rather than a load-bearing assertion -- said so in its own docstring rather than left to look like coverage. ## Not just deliver_files Six built-in tools declare an array or object parameter and were all exposed: | tool | parameter | |---|---| | `deliver_files` | `files` | | `ask_user` | `questions` | | `image_generate` | `images` | | `message` | `media` | | `create_playbook` | `skills` | | `video_generate` | `params` | The same failure was already seen from the other side: `acp_client/acp_agent.py` records `update_task_state` and `operations should be array` from a real dispatch. ## Verification ``` make lint-python ruff check + ruff format --check, 1850 files make lint-imports 9 contracts kept, 0 broken pytest tests/ 19,699 passed, 7 failed (all pre-existing, see below) ``` - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [ ] User-facing docs or screenshots -- no user-facing string **End to end, through the real tool**, with the exact payload that failed five times in production: ``` files sent as: str -> [{"path": ".../poster.png", "title": "..."}] returned: Delivered 1 file(s): poster.png (72 B). still "should be array": False registered in the store: True ``` Twelve cases, five mutations, each killing at least one: | mutation | red | |---|---| | drop the whole branch (the old behaviour) | 3 | | raise instead of falling through on a parse failure | 5 | | accept the parse whatever type it yielded | 1 | | do not re-cast the parsed value, so items are not cast | 1 | | widen the tuple to include `string` | **survived -- unreachable** | The last is worth stating rather than hiding: that mutation cannot change behaviour, because the `type == "string"` branch returns earlier and the type check would refuse the result anyway. I also tried moving the new block above that branch, alone and combined with widening the tuple, and neither reddens anything either. The invariant holds by construction; the case that documents it says so. **Pre-existing failures, checked against the base.** 7 failed in the full run, all in `tests/test_security_network.py` (`test_a_redirect_to_a_dangerous_address_is_refused` and siblings). Running that file alone on this same tree gives 4 -- the count is order-dependent -- and the change touches only `raven/agent/tools/params.py` plus a new test file, neither of which `raven/security/network.py` imports. The failures are the box's resolver: it returns poisoned answers roughly half the time, which I measured separately (`en.wikipedia.org` came back as `2001::1` on 3 of 6 lookups). Not introduced here, and worth someone's attention on its own. ## Risk - [x] Security impact considered -- widens what reaches a tool's own validation, not what bypasses it. A parsed value is still validated against the same schema, and a parse that yields the wrong type is refused exactly as before - [x] Backward compatibility considered -- a call that worked before takes the identical path; the new branch is reached only by a `str` where an array or object is declared, which previously always failed - [x] Rollback path is clear -- single-commit revert ## Related Issues N/A ## Swept `mr-review-patterns` pre-submit sweep, over `git diff origin/main...HEAD`. - **1 (a write location moved)** -- nothing moved; a branch was added to an existing dispatch on `target_type`. Grepped every caller of `cast_params`: the registry is the only one, and it runs `validate_params` right after, which is the reader this affects. - **2 (a value with several entry points)** -- the cast is the single funnel every tool argument passes through, which is why the fix belongs here and not in `deliver_files`. - **3 (a guard)** -- checked both directions: what it wrongly lets through (JSON of the wrong type, non-JSON text -- both refused, with cases) and what it wrongly refuses (a real list, unaffected, with a case). - **4 (tests)** -- five mutations above, one honestly reported as unreachable. Failing-ID sets diffed against the base rather than counts. - **6 (stale text)** -- the module docstring enumerated what is coerced ("a string that should be an integer"); it now names arrays and objects too. Grepped the tracked tree for other statements of the rule: none. - **9 (error handling)** -- the new path adds one way to end and it is caught: `json.loads` raises `ValueError` on bad text and `TypeError` on a non-string, and both fall through to the untouched value. - **11** -- `git ls-files | grep -c node_modules` is 0; the commit-message gate was run, not eyeballed. Not applicable, with reasons: **3b, 3c, 3d, 3e** (no tool schema produced, no state moved, one lane, no assertion lost its symbol); **4b, 4c, 4d** (no platform branch, no shell, no rendered number); **5, 5b** (no dependency; `json` is stdlib and already imported across this package); **6b, 6c** (no backup, no generated artifact); **7, 7b** (nothing on disk); **8** (no asset, no secret); **10** (no prompt text). ## Deliberately not fixed The error text still ends with `[Analyze the error above and try a different approach.]`, and the model plainly did not -- it sent the same shape five times. A message naming the expected spelling (`files must be a JSON array, not a string containing one`) would give it something to act on. That is a change to the registry's error surface, affects every tool, and deserves its own argument. ## Moved from !594 Opened on a compliant source branch. AGENTS.md 2.1 asks for a 3-5 word snake_case description and the previous branch had seven; GitLab cannot change an existing merge request's source branch, so the change moves rather than the name. Same commits, rebased onto `c2a9768f`. Both code findings from the review on !594 are already in them -- `RecursionError` caught with its regression case, and the implementation comment reduced to the two non-obvious reasons. Nothing else changed, so that review carries.
2026-09-09 15:42:36 +00:00
assert StructuredTextParser.supported_media_types == ["text/markdown", "text/html"]
left_behind = set(TextParser.supported_media_types) - set(StructuredTextParser.supported_media_types)
assert left_behind, "TextParser would be redundant, and the service could drop it"
assert "text/plain" in left_behind
def test_html_that_omits_the_head_close_is_still_sectioned():
"""`</head>` is optional in HTML and `html.parser` never closes it, so the
drop counter used to stay above zero for the rest of the file: every body came
out empty, no sections were produced, and `parse` fell back to one whole-file
Section carrying the raw markup -- the exact input this module exists to avoid.
"""
html = (
"<html>\n<head><title>Quarterly</title><style>body{color:red}</style>\n"
"<body>\n<h1>Findings</h1><p>Revenue rose.</p>\n"
"<h2>Risks</h2><p>Supply chain.</p>\n</body></html>"
)
sections = _parse(html, "q.html")
assert [s.metadata["heading"] for s in sections] == ["Findings", "Risks"]
body = "\n".join(s.content.text for s in sections)
assert "<style" not in body and "<title" not in body
assert "Revenue rose." in body and "Supply chain." in body
def test_a_script_inside_the_body_is_still_dropped():
"""The counter reset keys on `<body>`, which must not weaken the guard for
tags that legitimately open after it."""
html = "<html><body><h1>A</h1><script>steal()</script><p>one</p></body></html>"
sections = _parse(html, "q.html")
body = "\n".join(s.content.text for s in sections)
assert "one" in body
assert "steal()" not in body
feat(knowledge): read every format, show what the index holds, and let an agent search it (#489) <img width="2318" height="1448" alt="image" src="https://github.com/user-attachments/assets/bca71d69-4972-420c-b14e-c8cf74a17de9" /> ## Summary Everything a knowledge base could not read, and the pieces it cut them into saying what they are. **Reading a PDF.** A PDF records glyphs at coordinates; whether they are a heading, a table cell or a running footer is inferred from how they look. There are now two readers. The geometric one infers structure from type size and position and needs nothing but the file. The other asks deepdoc's models, ported into `raven/knowledge/parser/deepdoc`: a layout model that says what each region is, a detector and recognizer that read a page carrying no text at all, and a table-structure model that rebuilds a grid from a picture of one. Detecting boxes around text the file already spells out exactly costs an inference pass per page to learn less, and recognising it costs accuracy. The layout graph answers with a shape its own postprocessing raises on, and its classes are DocLayout-YOLO's order rather than the list Both were established by running the model, and the label order is pinned by a test. **Tables go in as tables.** `<table>` is the only shape that keeps what a grid states -- which column a cell stood under, which cells are merged, which row is the header -- and it is the shape a model reads a table in, which is what these chunks are for. PDF, Word and PowerPoint all state their merges exactly, and they now survive as `colspan` and `rowspan`. The flat form this replaces had to repeat a header standing over two columns across both of them, because there was nowhere to record a span; it had no callers left afterwards and is gone. Two deviations from upstream, both about what the string is for: every cell's text is escaped, where upstream interpolates it raw and a cell holding a `<` produces markup that is no longer a table, and the span attributes are written only where there is a span, where upstream emits `<td >` for every ordinary cell. **A picture of where each chunk came from.** A chunk is a few hundred words with a page number on it, which is enough to cite and not enough to answer whether the parser read the page correctly. So every PDF chunk that knows where it sits carries one image: the regions it covers, cropped out of the rendered pages and stacked. here they are already structured in the parser's element spans, so the markers would only put text nobody wrote into the thing being embedded. The panel draws a chunk's tables as tables, parsed and rebuilt as elements rather than injected: a person can rewrite any piece from that same panel, so what comes back from the store is user input however it got there. **The page that reads them.** Upstream retired the knowledge page while this branch was out; it is rebuilt against the new architecture, and the half that never existed is the one a reader needs -- what a document actually became once it was cut and embedded. Documents sit in one-level folders and arrive by four routes: a file, a folder, a URL, a written note. Adding and indexing are two calls on the contract and nothing drains that queue on its own, so a page that stopped after the first left every upload saying Queued for good. Opening a file takes the whole page: the document on the left, the pieces the index holds of it on the right. The left half is the file drawn as its own pages rather than framed by the browser's PDF viewer, which reads where to open once and answers to nothing afterwards -- and which nothing can be drawn on. Pages are rendered once and kept, and the region a piece was cut from is washed over the page it came from, placed in fractions of the page so the picture's own scale never enters the arithmetic. Clicking a piece takes the file there; double-clicking opens it for rewriting. The pieces can be searched, filtered by state, turned on and off, deleted, rewritten and added to, one at a time or in a batch. Every one of those was already on the contract with no reader for it. A base's settings are back and fuller: how many pieces a search returns, how documents are cut, and how much prose comes with a table or a figure. The embedding model is a picker while the base holds no pieces and a stated fact once it does -- moving it then is the collection made again at the new width and every document re-cut, which is not a thing a panel does by being saved. **An agent can read them.** The retrieval engine was there and nothing could reach it: no agent had a tool that opened a library. A reader now points a conversation at bases from the composer, the way they pin a working directory, and the selection rides in the same place -- the session's metadata, bound around the turn. The tool takes a query and never a base. The scope is the reader's choice, so a model cannot reach material nobody offered it, and it is withheld where nothing was picked rather than offered and refused. A hit says which document and page it came from, because a hit a reader cannot place is a hit they cannot check -- and the passages are pressable in the transcript, walking into the base, the document, the folder it is filed under and then the piece. A citation a reader cannot follow is a claim they have to take on trust, and the whole point of searching their own material is that they can go and look. **The word segmenter and the resources.** `raven.core.Tokenizer` is Infinity's `rag_tokenizer`, ported: a forward and backward maximum match over a dictionary trie, with stemming and lemmatisation on the latin path. XGBoost is not carried over. None of the data lives in git -- the dictionary, the NLTK corpora and the ONNX graphs run to about a hundred megabytes, so `make fetch-resources` on an install and an optional Docker layer behind `RAVEN_RESOURCES` both point at one script. A checkout that could not reach the network still works, and every feature that reads this data says so when asked. ## Type - [ ] Fix - [x] Feature - [ ] Docs - [ ] CI / tooling - [ ] Refactor - [ ] Other ## Verification Run on macOS against Python 3.12, with the deepdoc weights fetched (the model tests skip without them): ``` uv run pytest 27450 passed, 111 skipped, 3 failed npm run test --prefix ui-web 3276 passed npm run test --prefix ui-tui 2106 passed npm run type-check --prefix ui-web clean (tsc) npm run lint --prefix ui-web clean (4 pre-existing hook warnings) make lint-python clean (ruff check + format) make lint-types clean (ty) make lint-imports 10 contracts kept, 0 broken make check-source-language clean make check-large-files clean make build-ui built, both boot snapshots match node ui-web/scripts/check-css.mjs clean node ui-web/scripts/check-class-namespace.mjs clean ``` The three Python failures are not this branch's and were each confirmed on the base alone: one asks a host without LibreOffice to refuse a render and this host has LibreOffice installed, and two simulation cases fail on an unmodified checkout of the base commit. The RPC contract was regenerated for both consumers and both match it (`npm run gen --prefix ui-web`, `npm run gen:rpc --prefix ui-tui`, then `gen:check`). The page's boot snapshots and region goldens were refreshed where the composer bar gained a button, and each records that button and nothing else. - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [x] User-facing docs or screenshots are updated when needed ## Risk A PDF's tables, and a Word or PowerPoint file's, are now indexed as HTML rather than as header-prefixed lines. Chunk text changes shape for those documents, so a base indexed before this and searched after it holds both forms until its documents are reindexed; retrieval keeps working either way, because both forms are text. The same applies to how much prose a table or a figure carries with it: the setting now reaches as far as it says, where before it stopped at the nearest paragraph, so a chunk cut before this keeps the narrower context until its document is rebuilt. Rolling back is reverting the branch and reindexing. The deepdoc models are optional. An install that never ran `make fetch-resources` reads PDFs geometrically, exactly as before, and says so when the models are asked for by name. Nothing here reaches the network at request time. Chunk crops and rendered pages are written at index time and on first view, both under the state directory rather than in the repository, and both are deleted with their document. A page is immutable for the life of its document id, so the route serving them caches; the document route does not. All three routes take ids and no path, and the two image routes answer under the strictest sandbox in the tree. What an agent may search is the reader's choice and nothing else. The tool takes no base argument, so a model cannot name one, and it is absent from the schema entirely for a conversation that named none -- a lane nobody pointed at a base is unchanged by this branch. The bases a session carries are stored beside its working directory and travel no further. - [x] Security impact considered - [x] Backward compatibility considered - [x] Rollback path is clear for risky changes ## Related Issues N/A --------- Co-authored-by: Claude (claude-opus-5) <noreply@anthropic.com>
2026-10-12 19:05:16 +08:00
# -- cutting where the document allows -----------------------------
def _with_elements(parts: list[str]) -> Section:
"""A section whose element spans say where its parts begin and end.
What a parser records when it knows: the docx parser writes one span per
paragraph, table and figure. Built by hand here so the rule can be tested
without a document format in the way.
"""
text = "\n\n".join(parts)
spans = []
at = 0
for index, part in enumerate(parts):
spans.append({"reading_order": index, "layout_type": "text", "char_start": at, "char_end": at + len(part)})
at += len(part) + 2
return Section(content=TextBlock(text=text), source="handbook.docx", metadata={"elements": spans})
def test_a_paragraph_is_never_cut_in_half():
"""A paragraph is the unit a person wrote and the unit a reader reads. Half
of one is embedded as the half-thought it has become, and reads back as a
sentence that stops."""
paragraphs = [f"Paragraph {n}. " + " ".join(f"word{i}" for i in range(60)) for n in range(6)]
section = _with_elements(paragraphs)
chunks = _chunk([section], chunk_size=200, overlap=20)
assert len(chunks) > 1, "the section is too big for one chunk, so this is a real test"
for paragraph in paragraphs:
assert sum(paragraph in chunk.content.text for chunk in chunks) == 1, paragraph[:20]
def test_chunks_are_packed_with_whole_elements():
"""Packed until the next one will not fit, so a chunk is as full as whole
paragraphs allow rather than as full as bytes allow."""
section = _with_elements(["First.", "Second.", "Third."])
chunks = _chunk([section], chunk_size=200, overlap=20)
assert _texts(chunks) == ["First.\n\nSecond.\n\nThird."]
def test_an_element_larger_than_the_budget_is_still_split():
"""One paragraph can be longer than any window. Refusing to cut it would
mean refusing to index it."""
long_one = " ".join(f"word{i}" for i in range(400))
section = _with_elements([long_one, "A short one after it."])
chunks = _chunk([section], chunk_size=100, overlap=10)
assert len(chunks) > 1
assert "".join(c.content.text for c in chunks).count("word399") == 1
def test_a_split_inside_a_paragraph_falls_between_words():
"""Where it has to cut, it cuts at a break rather than through a word: two
halves of `paragraph` are not a word either half of the index knows."""
section = _with_elements([" ".join(f"word{i}" for i in range(400))])
chunks = _chunk([section], chunk_size=100, overlap=10)
for chunk in chunks[:-1]:
assert chunk.content.text.endswith(" ") or chunk.content.text[-1].isalnum()
# The tail of every piece but the last is a whole token.
tail = chunk.content.text.rstrip().rsplit(" ", 1)[-1]
assert tail.startswith("word") and tail[4:].isdigit(), tail
def test_a_section_with_no_element_spans_is_cut_as_before():
"""Markdown and plain text carry no spans, and nothing about them
changed."""
section = Section(content=TextBlock(text="x" * 4000), source="notes.md", metadata={})
chunks = _chunk([section], chunk_size=100, overlap=10)
assert len(chunks) > 1
def _tokens(text: str) -> int:
"""The same estimate the chunker sizes with."""
return len(text.encode("utf-8")) // 4
def test_the_heading_prefix_is_paid_out_of_the_chunk_budget():
"""The prefix is prepended to every piece after the first, so the split has
to be made against what is left after it. Split at the full size instead,
every later chunk comes back over the limit the budget exists to keep --
which is an embedding input limit, not a preference."""
path = ["Handbook", "Operations", "Escalation policy"]
section = Section(
content=TextBlock(text=" ".join(f"word{n}" for n in range(400))),
source="handbook.docx",
metadata={"heading_path": path},
)
chunks = _chunk([section], chunk_size=64, overlap=8)
assert len(chunks) > 1, "the section has to be split for this to mean anything"
assert max(_tokens(chunk.content.text) for chunk in chunks) <= 64
def test_an_element_too_big_on_its_own_is_split_against_the_budget_too():
"""The other branch, and the same rule: one paragraph longer than any
window still has to come back in pieces that fit with the prefix on."""
section = _with_elements([" ".join(f"word{n}" for n in range(400))])
section.metadata["heading_path"] = ["Handbook", "Operations", "Escalation policy"]
chunks = _chunk([section], chunk_size=64, overlap=8)
assert len(chunks) > 1
assert max(_tokens(chunk.content.text) for chunk in chunks) <= 64
def test_a_section_with_no_heading_still_cuts_at_the_full_size():
"""Nothing is prefixed to it, so nothing is taken off its budget."""
section = Section(content=TextBlock(text=" ".join(f"word{n}" for n in range(400))), source="n.md", metadata={})
chunks = _chunk([section], chunk_size=64, overlap=8)
assert max(_tokens(chunk.content.text) for chunk in chunks) <= 64
assert len(chunks) < 20, "and is not cut finer than it was asked to be"