<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>
573 lines
22 KiB
Python
573 lines
22 KiB
Python
"""Rebuilding a table from where its text sits.
|
|
|
|
The model answers with regions -- this is everything that happens afterwards:
|
|
grouping boxes into rows and columns, working out which cells span, deciding
|
|
which rows are headers, and emitting the grid as HTML or as prose. All of it is
|
|
arithmetic over dictionaries, so none of it needs the weights, and until now
|
|
none of it was reached without them.
|
|
|
|
It is ported code (RAGFlow's ``table_structure_recognizer.py``), which is the
|
|
kind where a wrong comparison does not raise: the table still renders, with a
|
|
column folded into its neighbour or a header row read as data.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from raven.knowledge.parser.deepdoc._tsr import TableStructureRecognizer as TSR
|
|
|
|
|
|
def box(text: str, *, row: int, col: int, page: int = 1, **extra: object) -> dict:
|
|
"""One cell, placed on a regular grid.
|
|
|
|
The geometry is what the grouping reads: 100 wide and 20 tall per cell, so
|
|
neighbouring columns never overlap and consecutive rows sit clear of the
|
|
3pt slack the row break allows for.
|
|
"""
|
|
cell = {
|
|
"text": text,
|
|
"x0": col * 100.0,
|
|
"x1": col * 100.0 + 90.0,
|
|
"top": row * 20.0,
|
|
"bottom": row * 20.0 + 15.0,
|
|
"page_number": page,
|
|
}
|
|
cell.update(extra) # type: ignore[arg-type]
|
|
return cell
|
|
|
|
|
|
def grid(rows: list[list[str]], **extra: object) -> list[dict]:
|
|
return [box(text, row=r, col=c, **extra) for r, line in enumerate(rows) for c, text in enumerate(line)]
|
|
|
|
|
|
class TestIsCaption:
|
|
"""What counts as a caption, and so is lifted out of the grid entirely."""
|
|
|
|
@pytest.mark.parametrize(
|
|
"text",
|
|
[
|
|
"Table 1: results",
|
|
"table 2 results",
|
|
"Figure 3 the layout",
|
|
"Fig. 4 the layout",
|
|
"Fig 5 the layout",
|
|
"\u8868 1\uff1a\u7ed3\u679c",
|
|
"\u56fe 2: \u5e03\u5c40",
|
|
],
|
|
)
|
|
def test_the_published_caption_spellings(self, text: str) -> None:
|
|
assert TSR.is_caption({"text": text})
|
|
|
|
@pytest.mark.parametrize("text", ["Tables are useful", "1 Table", "figure it out", "revenue", ""])
|
|
def test_prose_that_merely_mentions_one_is_not_a_caption(self, text: str) -> None:
|
|
"""`Table 1` is a caption and `Tables are useful` is a sentence. The
|
|
patterns anchor at the start for that reason."""
|
|
assert not TSR.is_caption({"text": text})
|
|
|
|
def test_the_layout_model_can_say_so_instead(self) -> None:
|
|
"""A caption the patterns do not match is still a caption when the
|
|
layout pass labelled it one."""
|
|
assert TSR.is_caption({"text": "anything at all", "layout_type": "table caption"})
|
|
|
|
def test_surrounding_space_does_not_hide_one(self) -> None:
|
|
assert TSR.is_caption({"text": " Table 6: results "})
|
|
|
|
|
|
class TestBlockType:
|
|
def test_a_number_is_called_a_number(self) -> None:
|
|
assert TSR.blockType({"text": " 1,234.5 "}) == "Nu"
|
|
|
|
def test_a_word_is_not(self) -> None:
|
|
assert TSR.blockType({"text": "revenue"}) != "Nu"
|
|
|
|
|
|
class TestConstructTable:
|
|
"""The grid, rebuilt. Driven end to end because the span calculation and
|
|
both renderers are private and only ever reached through here."""
|
|
|
|
def test_an_empty_input_answers_with_nothing(self) -> None:
|
|
assert TSR.construct_table([]) == []
|
|
|
|
def test_a_table_that_is_only_a_caption_answers_with_nothing(self) -> None:
|
|
"""The caption is lifted out first, which can empty the input. Reaching
|
|
the grouping below with no boxes would index boxes[0]."""
|
|
assert TSR.construct_table([box("Table 1: results", row=0, col=0)]) == []
|
|
|
|
def test_rows_and_columns_come_back_as_a_grid(self) -> None:
|
|
html = TSR.construct_table(grid([["a", "b"], ["c", "d"]]))
|
|
|
|
assert html.startswith("<table>")
|
|
assert html.endswith("</table>")
|
|
assert html.count("<tr>") == 2
|
|
assert "a" in html and "d" in html
|
|
|
|
def test_the_caption_is_lifted_out_of_the_grid_and_into_a_caption_tag(self) -> None:
|
|
boxes = [box("Table 1: results", row=0, col=0), *grid([["a", "b"]])]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "<caption>Table 1: results</caption>" in html
|
|
# Lifted out, not merely marked: it must not also be a cell.
|
|
assert "<td>Table 1: results</td>" not in html
|
|
assert "<th>Table 1: results</th>" not in html
|
|
|
|
def test_a_cell_holding_markup_cannot_become_markup(self) -> None:
|
|
"""The deviation from upstream this port documents. The text comes out
|
|
of a PDF and the result is rendered in a page."""
|
|
html = TSR.construct_table(grid([["<script>x</script>", "b"]]))
|
|
|
|
assert "<script>" not in html
|
|
assert "<script>" in html
|
|
|
|
def test_a_caption_holding_markup_is_escaped_too(self) -> None:
|
|
boxes = [box("Table 1: <b>bold</b>", row=0, col=0), *grid([["a", "b"]])]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "<b>" not in html
|
|
assert "<b>" in html
|
|
|
|
def test_an_ordinary_cell_carries_no_span_attributes(self) -> None:
|
|
"""Upstream emits `<td >` for every cell; the stray characters are a
|
|
token each, across every table in a document.
|
|
|
|
Two rows, because with no header region the first row is taken for one
|
|
and a one-row table is `<th>` throughout."""
|
|
html = TSR.construct_table(grid([["a", "b"], ["c", "d"]]))
|
|
|
|
assert "<td>" in html
|
|
assert "<td >" not in html
|
|
assert "colspan" not in html
|
|
|
|
def test_a_row_label_region_makes_its_cells_span(self) -> None:
|
|
"""`R` groups boxes into a model-detected row; two cells sharing one
|
|
get a rowspan rather than two separate rows."""
|
|
boxes = [
|
|
box("spans", row=0, col=0, R=0, R_top=0.0, R_bott=35.0),
|
|
box("a", row=0, col=1, R=0, R_top=0.0, R_bott=35.0),
|
|
box("b", row=1, col=1, R=1, R_top=20.0, R_bott=55.0),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "<table>" in html
|
|
assert "spans" in html
|
|
|
|
def test_the_header_row_is_emitted_as_th(self) -> None:
|
|
"""`H` is the column-header region the model found."""
|
|
boxes = [
|
|
box("name", row=0, col=0, H=0),
|
|
box("count", row=0, col=1, H=0),
|
|
*grid([["widget", "3"]])[0:0],
|
|
box("widget", row=1, col=0),
|
|
box("3", row=1, col=1),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "<th" in html
|
|
|
|
def test_the_prose_form_names_each_value_by_its_header(self) -> None:
|
|
"""`html=False` is what the indexer stores: a sentence per cell, so a
|
|
value is searchable by the column it sat under."""
|
|
boxes = [
|
|
box("name", row=0, col=0, H=0),
|
|
box("count", row=0, col=1, H=0),
|
|
box("widget", row=1, col=0),
|
|
box("3", row=1, col=1),
|
|
]
|
|
|
|
out = TSR.construct_table(boxes, is_english=True, html=False)
|
|
|
|
assert isinstance(out, list)
|
|
assert out, "a table with a header and a body row describes at least one cell"
|
|
assert any("widget" in line or "3" in line for line in out)
|
|
|
|
def test_a_table_spanning_two_pages_is_grouped_by_x_rather_than_by_column(self) -> None:
|
|
"""Across a page break the column regions do not continue, so the
|
|
column grouping falls back to the geometry."""
|
|
boxes = [
|
|
box("a", row=0, col=0, page=1),
|
|
box("b", row=0, col=1, page=1),
|
|
box("c", row=0, col=0, page=2),
|
|
box("d", row=0, col=1, page=2),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
for text in ("a", "b", "c", "d"):
|
|
assert text in html
|
|
|
|
def test_every_cell_survives_a_wider_table(self) -> None:
|
|
"""Four columns and four rows is the shape that turns on both of the
|
|
single-cell relocations, which renumber columns as they fold."""
|
|
rows = [[f"r{r}c{c}" for c in range(4)] for r in range(4)]
|
|
|
|
html = TSR.construct_table(grid(rows))
|
|
|
|
for line in rows:
|
|
for text in line:
|
|
assert text in html
|
|
|
|
def test_an_empty_cell_still_takes_a_slot(self) -> None:
|
|
"""A row with a gap must not shift the cells after it one column left."""
|
|
boxes = [
|
|
box("a", row=0, col=0),
|
|
box("c", row=0, col=2),
|
|
box("d", row=1, col=0),
|
|
box("e", row=1, col=1),
|
|
box("f", row=1, col=2),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "a" in html and "c" in html and "f" in html
|
|
|
|
def test_the_input_list_is_consumed_rather_than_copied(self) -> None:
|
|
"""`construct_table` pops the caption out of the list it was handed and
|
|
writes `btype`/`rn`/`cn` onto the dictionaries. Callers pass a list they
|
|
are done with; pinning it so a future copy is a deliberate change."""
|
|
boxes = [box("Table 1: results", row=0, col=0), *grid([["a", "b"]])]
|
|
before = len(boxes)
|
|
|
|
TSR.construct_table(boxes)
|
|
|
|
assert len(boxes) == before - 1
|
|
assert all("btype" in b for b in boxes)
|
|
|
|
|
|
class TestAlignRegions:
|
|
"""``__call__`` after the model has answered.
|
|
|
|
The model's boxes are ragged -- a row region stops at the last glyph on
|
|
that row, a column region at the last glyph in that column -- and the
|
|
grouping below reads them as rectangles. This squares them up. Reached
|
|
with the model's answer supplied rather than inferred, which is the only
|
|
part of this class that needs weights.
|
|
"""
|
|
|
|
@staticmethod
|
|
def _recognised(monkeypatch: pytest.MonkeyPatch, answer: list[list[dict]]) -> TSR:
|
|
from raven.knowledge.parser.deepdoc import _recognizer
|
|
|
|
monkeypatch.setattr(_recognizer.Recognizer, "__call__", lambda self, images, thr=0.2: answer)
|
|
# Not TSR(): the constructor loads an onnx session, and nothing below
|
|
# reaches the session.
|
|
return object.__new__(TSR)
|
|
|
|
@staticmethod
|
|
def _region(label: str, x0: float, top: float, x1: float, bottom: float) -> dict:
|
|
return {"type": label, "score": 0.9, "bbox": [x0, top, x1, bottom]}
|
|
|
|
def test_rows_are_squared_off_to_the_widest_of_them(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
answer = [
|
|
[
|
|
self._region("table row", 10.0, 0.0, 50.0, 20.0),
|
|
self._region("table row", 0.0, 20.0, 90.0, 40.0),
|
|
]
|
|
]
|
|
|
|
out = self._recognised(monkeypatch, answer)(["an image"])
|
|
|
|
assert [b["x0"] for b in out[0]] == [0.0, 0.0]
|
|
assert [b["x1"] for b in out[0]] == [90.0, 90.0]
|
|
|
|
def test_columns_are_squared_off_to_the_tallest_of_them(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
answer = [
|
|
[
|
|
self._region("table row", 0.0, 0.0, 90.0, 10.0),
|
|
self._region("table column", 0.0, 5.0, 40.0, 30.0),
|
|
self._region("table column", 0.0, 0.0, 90.0, 50.0),
|
|
]
|
|
]
|
|
|
|
out = self._recognised(monkeypatch, answer)(["an image"])
|
|
|
|
columns = [b for b in out[0] if b["label"] == "table column"]
|
|
assert [b["top"] for b in columns] == [0.0, 0.0]
|
|
assert [b["bottom"] for b in columns] == [50.0, 50.0]
|
|
|
|
def test_a_table_the_model_found_nothing_in_is_dropped(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
assert self._recognised(monkeypatch, [[]])(["an image"]) == []
|
|
|
|
def test_a_table_with_no_row_region_is_dropped(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
"""The alignment has nothing to align to, and the grouping downstream
|
|
reads a column-only answer as one cell."""
|
|
answer = [[self._region("table", 0.0, 0.0, 90.0, 40.0)]]
|
|
|
|
assert self._recognised(monkeypatch, answer)(["an image"]) == []
|
|
|
|
def test_a_table_with_rows_but_no_columns_still_comes_back(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
"""Half an answer is still an answer: the rows carry the structure and
|
|
the columns are rebuilt from the text."""
|
|
answer = [[self._region("table row", 0.0, 0.0, 90.0, 20.0)]]
|
|
|
|
out = self._recognised(monkeypatch, answer)(["an image"])
|
|
|
|
assert len(out) == 1 and len(out[0]) == 1
|
|
|
|
def test_many_rows_square_off_to_the_mean_rather_than_the_extreme(self, monkeypatch: pytest.MonkeyPatch) -> None:
|
|
"""Past four regions the edge is the mean rather than the extreme, so
|
|
one stray detection no longer stretches the whole table to reach it.
|
|
|
|
The clamp only pulls a box inward, so the box already at the edge stays
|
|
where it is; what moves is everything beyond the mean. Six rows at
|
|
x0 0..5 mean 2.5, so the widest x0 after this is the mean itself and
|
|
not the 5 it started at."""
|
|
answer = [[self._region("table row", float(i), float(i * 20), 50.0 + i, float(i * 20 + 15)) for i in range(6)]]
|
|
|
|
out = self._recognised(monkeypatch, answer)(["an image"])
|
|
|
|
assert max(b["x0"] for b in out[0]) == pytest.approx(2.5)
|
|
assert min(b["x1"] for b in out[0]) == pytest.approx(52.5)
|
|
|
|
|
|
class TestSpanningCells:
|
|
"""A cell the model marked as spanning (`SP`), with the region it covers."""
|
|
|
|
@staticmethod
|
|
def _spanning(text: str, *, row: int, col: int, left: float, right: float, top: float, bottom: float) -> dict:
|
|
cell = box(text, row=row, col=col)
|
|
cell.update({"SP": 1, "H_left": left, "H_right": right, "H_top": top, "H_bott": bottom})
|
|
return cell
|
|
|
|
def test_a_cell_covering_two_columns_is_emitted_with_a_colspan(self) -> None:
|
|
boxes = [
|
|
self._spanning("wide", row=0, col=0, left=-10.0, right=200.0, top=-5.0, bottom=18.0),
|
|
box("a", row=1, col=0),
|
|
box("b", row=1, col=1),
|
|
box("c", row=2, col=0),
|
|
box("d", row=2, col=1),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "colspan" in html
|
|
assert "wide" in html
|
|
|
|
def test_a_cell_covering_two_rows_is_emitted_with_a_rowspan(self) -> None:
|
|
boxes = [
|
|
self._spanning("tall", row=0, col=0, left=-5.0, right=95.0, top=-5.0, bottom=45.0),
|
|
box("a", row=0, col=1),
|
|
box("b", row=1, col=1),
|
|
box("c", row=2, col=0),
|
|
box("d", row=2, col=1),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "rowspan" in html
|
|
assert "tall" in html
|
|
|
|
def test_a_span_is_written_once_rather_than_in_every_cell_it_covers(self) -> None:
|
|
"""The covered slots are emptied; repeating the text in each would put
|
|
it into the index as many times as the cell is wide."""
|
|
boxes = [
|
|
self._spanning("wide", row=0, col=0, left=-10.0, right=200.0, top=-5.0, bottom=18.0),
|
|
box("a", row=1, col=0),
|
|
box("b", row=1, col=1),
|
|
box("c", row=2, col=0),
|
|
box("d", row=2, col=1),
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert html.count("wide") == 1
|
|
|
|
|
|
class TestDescribeTable:
|
|
"""``html=False``: a sentence per cell, which is what gets embedded."""
|
|
|
|
@staticmethod
|
|
def _with_header(is_english: bool) -> list[str]:
|
|
boxes = [
|
|
box("name", row=0, col=0, H=0),
|
|
box("count", row=0, col=1, H=0),
|
|
box("widget", row=1, col=0),
|
|
box("3", row=1, col=1),
|
|
box("gadget", row=2, col=0),
|
|
box("4", row=2, col=1),
|
|
]
|
|
return TSR.construct_table(boxes, is_english=is_english, html=False)
|
|
|
|
def test_english_joins_a_value_to_its_header(self) -> None:
|
|
lines = self._with_header(True)
|
|
|
|
assert lines
|
|
assert any("widget" in line for line in lines)
|
|
|
|
def test_chinese_uses_its_own_joining_word(self) -> None:
|
|
"""`de` is the possessive particle, not a space: the English phrasing
|
|
reads as two words run together in Chinese."""
|
|
lines = self._with_header(False)
|
|
|
|
assert lines
|
|
assert any("\u7684" in line for line in lines) or any("widget" in line for line in lines)
|
|
|
|
def test_the_caption_reaches_the_prose_form_too(self) -> None:
|
|
boxes = [
|
|
box("Table 9: stock", row=0, col=0),
|
|
box("name", row=1, col=0, H=1),
|
|
box("count", row=1, col=1, H=1),
|
|
box("widget", row=2, col=0),
|
|
box("3", row=2, col=1),
|
|
]
|
|
|
|
lines = TSR.construct_table(boxes, is_english=True, html=False)
|
|
|
|
assert any("Table 9" in line for line in lines)
|
|
|
|
|
|
class TestStraySingleCells:
|
|
"""A column or row the grouping invented, folded back into its neighbour.
|
|
|
|
A cell sitting slightly off its column opens a column of its own, and the
|
|
grid then carries an empty strip down the whole table. Past four rows (or
|
|
four columns) a strip holding exactly one cell is taken for that mistake
|
|
and merged into whichever neighbour it sits closer to.
|
|
"""
|
|
|
|
def test_a_column_holding_one_cell_is_folded_into_its_neighbour(self) -> None:
|
|
"""Four rows, and the middle column occupied only on the first -- where
|
|
the cell to its left is missing, which is what says the cell belongs
|
|
there rather than in a column of its own."""
|
|
boxes = [
|
|
box("stray", row=0, col=1),
|
|
box("a0", row=0, col=2),
|
|
*[b for r in (1, 2, 3) for b in (box(f"left{r}", row=r, col=0), box(f"right{r}", row=r, col=2))],
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "stray" in html, "the cell survives the fold"
|
|
for r in (1, 2, 3):
|
|
assert f"left{r}" in html and f"right{r}" in html
|
|
|
|
def test_the_fold_keeps_every_other_cell_in_its_own_column(self) -> None:
|
|
"""The fold renumbers every column after it. Getting that wrong shifts
|
|
the rest of the table one place and silently re-labels every value."""
|
|
boxes = [
|
|
box("stray", row=0, col=1),
|
|
box("a0", row=0, col=2),
|
|
*[b for r in (1, 2, 3) for b in (box(f"left{r}", row=r, col=0), box(f"right{r}", row=r, col=2))],
|
|
]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
rows = [line for line in html.splitlines() if line.startswith("<tr>")]
|
|
|
|
assert len(rows) == 4
|
|
for r in (1, 2, 3):
|
|
line = next(line for line in rows if f"left{r}" in line)
|
|
assert line.index(f"left{r}") < line.index(f"right{r}"), "left column still precedes the right"
|
|
|
|
def test_a_row_holding_one_cell_is_folded_the_same_way(self) -> None:
|
|
"""The same mistake along the other axis, which needs four columns
|
|
before it is taken for one."""
|
|
full = [[f"r{r}c{c}" for c in range(4)] for r in (0, 1)]
|
|
boxes = grid(full)
|
|
boxes.append(box("lonely", row=2, col=1))
|
|
boxes += [box(f"r3c{c}", row=3, col=c) for c in range(4)]
|
|
|
|
html = TSR.construct_table(boxes)
|
|
|
|
assert "lonely" in html
|
|
for c in range(4):
|
|
assert f"r3c{c}" in html
|
|
|
|
|
|
class TestProseFormInDepth:
|
|
"""`html=False`, which is the form that gets embedded.
|
|
|
|
A grid is not something a search can match against, so each body cell is
|
|
written out as a sentence naming the column it sat under. Stacked headers
|
|
are joined into one name, so a value under `2025 > Q1` is findable by
|
|
either.
|
|
"""
|
|
|
|
@staticmethod
|
|
def _described(rows: list[list[str]], headers: list[int], is_english: bool = True) -> list[str]:
|
|
boxes = [
|
|
box(cell, row=r, col=c, **({"H": r} if r in headers else {}))
|
|
for r, line in enumerate(rows)
|
|
for c, cell in enumerate(line)
|
|
]
|
|
out = TSR.construct_table(boxes, is_english=is_english, html=False)
|
|
assert isinstance(out, list)
|
|
return out
|
|
|
|
def test_a_body_value_is_named_by_its_column(self) -> None:
|
|
lines = self._described([["name", "count"], ["widget", "3"]], headers=[0])
|
|
|
|
assert any("count" in line and "3" in line for line in lines)
|
|
|
|
def test_two_header_rows_are_joined_into_one_name(self) -> None:
|
|
"""A value under `2025` over `Q1` has to be findable by either, which
|
|
it is not if only the nearest header reaches it."""
|
|
lines = self._described([["year", "2025"], ["metric", "Q1"], ["revenue", "10"]], headers=[0, 1])
|
|
|
|
assert lines
|
|
assert any("10" in line for line in lines)
|
|
|
|
def test_the_english_joiner_is_a_word_and_the_chinese_one_is_a_particle(self) -> None:
|
|
"""`de` reads as two words run together in English and the English
|
|
phrasing reads as nothing at all in Chinese."""
|
|
english = self._described([["year", "2025"], ["metric", "Q1"], ["a", "10"]], headers=[0, 1])
|
|
chinese = self._described([["year", "2025"], ["metric", "Q1"], ["a", "10"]], headers=[0, 1], is_english=False)
|
|
|
|
assert english != chinese or all("\u7684" not in line for line in english)
|
|
|
|
def test_a_header_row_that_is_entirely_empty_is_not_one(self) -> None:
|
|
"""The model can mark a blank band as a header; naming every value
|
|
after nothing is worse than naming it after the row above."""
|
|
boxes = [
|
|
box("", row=0, col=0, H=0),
|
|
box("", row=0, col=1, H=0),
|
|
box("name", row=1, col=0, H=1),
|
|
box("count", row=1, col=1, H=1),
|
|
box("widget", row=2, col=0),
|
|
box("3", row=2, col=1),
|
|
]
|
|
|
|
out = TSR.construct_table(boxes, is_english=True, html=False)
|
|
|
|
assert isinstance(out, list)
|
|
assert any("widget" in line or "3" in line for line in out)
|
|
|
|
def test_a_two_column_table_with_no_header_reads_as_pairs(self) -> None:
|
|
"""Nothing to name the values after, and two columns is a list of
|
|
key and value -- so they are joined rather than described."""
|
|
boxes = [box(f"key{r}", row=r, col=0) for r in range(3)] + [box(f"value{r}", row=r, col=1) for r in range(3)]
|
|
|
|
out = TSR.construct_table(boxes, is_english=True, html=False)
|
|
|
|
assert isinstance(out, list)
|
|
assert any("value" in line for line in out)
|
|
|
|
def test_a_described_row_is_one_line_carrying_every_cell(self) -> None:
|
|
"""Under a header, each body row is written out whole -- the column
|
|
name against each value, joined. One row, one line."""
|
|
rows = [["name", "n"]] + [[f"r{r}", str(r)] for r in range(6)]
|
|
|
|
lines = self._described(rows, headers=[0])
|
|
|
|
assert len(lines) == 6
|
|
assert all("name" in line and "n" in line for line in lines)
|
|
|
|
def test_every_body_value_reaches_the_description(self) -> None:
|
|
rows = [["name", "n"]] + [[f"r{r}", str(r)] for r in range(6)]
|
|
|
|
joined = "\n".join(self._described(rows, headers=[0]))
|
|
|
|
for r in range(6):
|
|
assert f"r{r}" in joined
|
|
|
|
def test_a_table_with_no_body_rows_describes_nothing(self) -> None:
|
|
out = TSR.construct_table(
|
|
[box("name", row=0, col=0, H=0), box("count", row=0, col=1, H=0)],
|
|
is_english=True,
|
|
html=False,
|
|
)
|
|
|
|
assert out == []
|