Raven-1/tests/test_knowledge_parser_pdfdoc.py
Kevin Hu 82c76a82a9
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

271 lines
9.8 KiB
Python

"""The one PDF reader: what a page says, what it draws, and what it looks like.
The fixtures are written with PyMuPDF, which is a dev dependency for exactly
this reason -- the reader under test can read a PDF but not draw one, and a
fixture has to be drawn before it can be read.
"""
from __future__ import annotations
import pytest
from raven.knowledge.parser._pdfdoc import (
Block,
Line,
PasswordProtectedError,
_grouped,
_lines_of,
open_document,
)
def _pdf(build) -> bytes:
"""A PDF built page by page, as bytes."""
import pymupdf
document = pymupdf.open()
build(document, pymupdf)
raw = document.tobytes()
document.close()
return raw
def _one(*, text: str = "Hello world", at: float = 100.0, size: float = 11.0) -> bytes:
def build(document, pymupdf):
document.new_page().insert_text(pymupdf.Point(60, at), text, fontsize=size)
return _pdf(build)
def _word(text: str, x0: float, x1: float, top: float = 0.0, bottom: float = 10.0, size: float = 10.0) -> dict:
return {"text": text, "x0": x0, "x1": x1, "top": top, "bottom": bottom, "size": size}
def _line(text: str, x0: float, top: float, x1: float, bottom: float) -> Line:
return Line(text=text, size=10.0, x0=x0, top=top, x1=x1, bottom=bottom)
# -- opening -------------------------------------------------------
def test_a_document_opens_from_bytes_and_from_a_path(tmp_path) -> None:
raw = _one()
path = tmp_path / "one.pdf"
path.write_bytes(raw)
with open_document(raw) as from_bytes, open_document(path) as from_path:
assert len(from_bytes) == len(from_path) == 1
assert from_bytes[0].blocks()[0].lines[0].text == from_path[0].blocks()[0].lines[0].text
def test_a_path_that_is_not_there_is_not_a_read_failure(tmp_path) -> None:
"""Told apart because a caller can do something about one of them."""
with pytest.raises(FileNotFoundError):
open_document(tmp_path / "gone.pdf")
def test_something_that_is_not_a_pdf_raises(tmp_path) -> None:
with pytest.raises(Exception) as caught: # noqa: B017 - the reader's own type, whatever it is
open_document(b"this is not a pdf at all")
assert not isinstance(caught.value, PasswordProtectedError)
def test_a_locked_file_is_told_apart_from_a_broken_one(tmp_path) -> None:
"""The reader wraps every cause in one type, so this is recovered from the
chain underneath it rather than from the exception itself."""
import pymupdf
document = pymupdf.open()
document.new_page().insert_text(pymupdf.Point(60, 80), "secret")
path = tmp_path / "locked.pdf"
document.save(str(path), encryption=pymupdf.PDF_ENCRYPT_AES_256, owner_pw="o", user_pw="u")
document.close()
with pytest.raises(PasswordProtectedError):
open_document(path)
def test_the_pages_are_numbered_from_one() -> None:
def build(document, pymupdf):
for index in range(3):
document.new_page().insert_text(pymupdf.Point(60, 100), f"page {index}", fontsize=11)
with open_document(_pdf(build)) as document:
assert document.page_count == len(document) == 3
assert [page.number for page in document] == [1, 2, 3]
assert document[1].number == 2
# -- what a page says ----------------------------------------------
def test_a_page_knows_how_big_it_is() -> None:
def build(document, pymupdf):
document.new_page(width=400, height=250)
with open_document(_pdf(build)) as document:
page = document[0]
assert (page.width, page.height) == (400.0, 250.0)
assert page.box == (0.0, 0.0, 400.0, 250.0)
def test_the_words_of_one_line_come_back_as_one_line() -> None:
with open_document(_one(text="Revenue grew in every region")) as document:
lines = [line for block in document[0].blocks() for line in block.lines]
assert [line.text for line in lines] == ["Revenue grew in every region"]
assert lines[0].size == 11.0
def test_a_line_is_cut_where_the_gap_is_too_wide_to_be_a_space() -> None:
"""Both columns of a two-column page share a baseline, so without this they
read straight across the gutter -- which puts the right-hand column's words
inside the left-hand column's region."""
close = _lines_of([_word("left", 60, 90), _word("column", 94, 130)])
apart = _lines_of([_word("left", 60, 90), _word("right", 300, 340)])
assert [line.text for line in close] == ["left column"]
assert [line.text for line in apart] == ["left", "right"]
def test_a_blocks_size_is_the_one_most_of_its_text_is_set_in() -> None:
"""Weighted by how much is set in it, not by how many runs there are: a
footnote marker is one run and the sentence around it is one run."""
(line,) = _lines_of([_word("a", 60, 64, size=22.0), _word("long sentence of body text", 66, 200, size=10.0)])
assert line.size == 10.0
def test_two_lines_of_one_paragraph_are_one_block() -> None:
first, second = _line("first line", 60, 100, 200, 110), _line("second line", 60, 112, 190, 122)
assert [[line.text for line in group] for group in _grouped([first, second])] == [["first line", "second line"]]
def test_a_paragraph_break_starts_a_new_block() -> None:
first, second = _line("a paragraph", 60, 100, 200, 110), _line("another", 60, 160, 190, 170)
assert len(_grouped([first, second])) == 2
def test_two_columns_are_two_blocks_though_their_lines_interleave() -> None:
"""The lines arrive down the page, so they alternate left, right, left: a
line compared only with the one before it never finds its own column."""
lines = [
_line(f"{side} {index}", x0, 100 + index * 12, x0 + 100, 110 + index * 12)
for index in range(3)
for side, x0 in (("left", 60.0), ("right", 300.0))
]
groups = _grouped(lines)
assert len(groups) == 2
assert {group[0].text for group in groups} == {"left 0", "right 0"}
assert all(len(group) == 3 for group in groups)
def test_a_picture_is_a_block_with_nothing_to_read() -> None:
def build(document, pymupdf):
page = document.new_page()
page.insert_text(pymupdf.Point(60, 100), "a caption", fontsize=11)
page.insert_image(
pymupdf.Rect(300, 400, 420, 500), pixmap=pymupdf.Pixmap(pymupdf.csRGB, pymupdf.IRect(0, 0, 20, 20))
)
with open_document(_pdf(build)) as document:
pictures = [block for block in document[0].blocks() if block.picture]
assert len(pictures) == 1
assert pictures[0].lines == ()
assert pictures[0].box == (300.0, 400.0, 420.0, 500.0)
def test_text_drawn_past_the_edge_of_the_page_is_not_read() -> None:
"""A PDF's content stream is not bounded by its page: a line set too long
carries on past the edge, and the reader never sees the rest. Read whole it
comes back as a box wider than the page, which is a box no region of that
page can contain -- so the layout model, which places text by asking which
region it fell in, drops it."""
with open_document(_one(text="Alpha " * 40, size=10.0)) as document:
page = document[0]
lines = [line for block in page.blocks() for line in block.lines]
assert lines, "the part that is printed is still read"
assert max(line.x1 for line in lines) <= page.width + 1
assert len(lines[0].text) < len("Alpha " * 40), "and the part that is not is not"
def test_a_page_holds_what_reading_it_cost_until_it_is_released() -> None:
with open_document(_one()) as document:
page = document[0]
first = page.blocks()
assert page.blocks() is first, "asked twice, assembled once"
page.release()
assert page.blocks() is not first, "and read again from scratch after"
assert [line.text for block in page.blocks() for line in block.lines] == ["Hello world"]
# -- tables --------------------------------------------------------
def test_a_ruled_table_is_found_with_its_cells() -> None:
def build(document, pymupdf):
page = document.new_page()
xs, ys = [60, 180, 300], [220, 250, 280]
for y in ys:
page.draw_line(pymupdf.Point(xs[0], y), pymupdf.Point(xs[-1], y))
for x in xs:
page.draw_line(pymupdf.Point(x, ys[0]), pymupdf.Point(x, ys[-1]))
for row in range(2):
for column in range(2):
page.insert_text(pymupdf.Point(xs[column] + 8, ys[row] + 20), f"r{row}c{column}", fontsize=11)
with open_document(_pdf(build)) as document:
tables = document[0].tables()
assert len(tables) == 1
assert tables[0].extract() == [["r0c0", "r0c1"], ["r1c0", "r1c1"]]
def test_a_page_whose_tables_cannot_be_found_still_reads_as_prose() -> None:
"""Table finding is heuristic and occasionally throws on a page whose
drawing commands it cannot make sense of. One page's tables are not worth
the document."""
class _Throws:
page_number = 1
def find_tables(self):
raise RuntimeError("cannot make sense of these drawings")
from raven.knowledge.parser._pdfdoc import Page
assert Page(_Throws(), 1).tables() == []
# -- what it looks like --------------------------------------------
def test_a_page_renders_at_the_scale_it_is_asked_for() -> None:
def build(document, pymupdf):
document.new_page(width=200, height=100)
with open_document(_pdf(build)) as document:
page = document[0]
for scale in (1, 3):
image = page.render(scale)
try:
assert image.size == (200 * scale, 100 * scale)
assert image.mode == "RGB", "the models were trained on three channels"
finally:
image.close()
def test_a_block_carries_the_box_around_every_line_in_it() -> None:
block = Block(lines=(), x0=1.0, top=2.0, x1=3.0, bottom=4.0)
assert block.box == (1.0, 2.0, 3.0, 4.0)
assert _line("x", 1.0, 2.0, 3.0, 4.0).box == (1.0, 2.0, 3.0, 4.0)