| Age | Commit message (Collapse) | Author | Files | Lines |
|
|
|
|
|
One reader where there were two. sisudoc-ssp-reader read a .ssp and
sisudoc-ocda-db-reader a .ocda.db, each with its own parser; this one
is built on the library, so every mode reads a .ssp, a .ocda.db and a
.sisupod, and says the same thing of each. Its output modules are the
.ssp reader's, copied (the reader repositories are unchanged) and
moved onto the library's model, which was the reader's types renamed
and added to. From the database reader, --languages and --files.
ocda-read [--tree|--outline|--html|--llm-text|--llm-jsonl|--images|
--verify --image-dir=<d>|--languages|--files|
--report-unknown|--validate] <file> [lang]
Two modes are about the file and say so: --files answers for a
.ocda.db, and --validate checks a .ssp as text and refuses anything
else. Where a file holds several languages and none is named, every
form reads the first and says so, where the database used to take one
and the pod refuse.
In src/read/ beside the library in src/ocda/; dub.json keeps it out of
the library (configurations "library" and "ocda-read"), and the flake
builds it as a package. The flake's default is now all three programs
in one bin/: built separately into the same result link, each replaced
the one before, and the pipeline's dump looked for ocda-dump where
only ocda-agree was left.
Checked over the 36 document-languages:
- against the .ssp reader, on a .ssp, byte for byte after the first
line (Parsed: is Read:): --tree, --html, --llm-text, --llm-jsonl,
--images, --verify, --validate, 36 of 36; the summary differs only
in the closing hint, which names the two new modes;
- against the database reader, on a .ocda.db: --files 27 of 27, and
--languages 27 of 27 on language, source and digest (its id column,
the database's own key, is left out);
- the same output from the .ssp, the .ocda.db and the .sisupod: the
summary and --tree, --html, --llm-text, --llm-jsonl, --images and
--report-unknown, 36 of 36; --languages, .ocda.db == .sisupod 27 of
27, and a single-language .ssp == its database, 26 of 26.
Not reproduced, deliberately: the database reader's --outline, which
ordered the headings by their position within their own section, so
that the toc and the "(skip)" headings of other sections fell among
the body's. --outline here is --tree.
(assisted by Claude-Code)
|
|
A pod built with --pod2 carries the abstraction inside it, at
media/abstraction/<uid>.<lang>.ssp. So ocda.load gains .sisupod and .zip,
and because the dispatch is in one place a writer over the library gained
a third input form without being touched.
Over the reference collection a .sisupod gives the same model as the .ssp
beside it: 36 of 36 document-languages, byte identical dumps. And the
text writer, unchanged, renders a document from its .sisupod byte
identically to spine --text, 36 of 36.
Two small readers, neither of them a library: ocda.zstd declares the four
libzstd entry points needed to unwrap a frame, as ocda.sqlite declares
sqlite3; ocda.zip reads a central directory and one stored entry. Deflate
is refused rather than mishandled, since spine writes stored entries and
a library that says it read what it skipped is worse than one that says
it cannot. A frame that declares an expansion over 512 MiB is refused
before anything is allocated: a reader of an artefact it did not make
declines to be told how much memory to take.
The Gleam library reads the zip and not the frame. The only zstd on the
BEAM is a NIF that clones zstd from github in its own build, so it cannot
be built sandboxed without patching it. A .zip pod is read in both
languages, a .sisupod in D only, and the Gleam refusal says why and says
"zstd -d" rather than claiming the file is malformed. It is a packaging
limit, not anything about the format, and the only place the two
libraries differ in what they accept.
A pod without an abstraction is refused with a message naming the flag
that puts one in, because that is a fact about the pod.
(assisted by Claude-Code)
|
|
The agreement check compared ten fields of each object. That was enough
to find the blank-first-line defect and not enough for anything
subtler, so this adds ocda-dump: the whole loaded model written out
canonically, every field, sorted dictionaries, sections in format
order, defaults omitted. It is an oracle, not an output format. Two
implementations of this model can now be compared with diff.
Run against the reference collection it found four disagreements
between the two loaders, none of which the ten-field check could see.
1. .identifier and .anchor. The .ssp emits a property only when it
carries information, so the identifier is absent when it is simply
the ocn and the anchor is absent when it is the segment name; the
database stores both unconditionally. Over one document that is
1,314 identifiers and 69 anchors. The loader now derives them, which
is what the model means: an object's identifier is its ocn unless it
says otherwise.
2. make and doc_has. The database has one metadata table where the
text form has three blocks, and tells them apart by a key prefix.
The loader now unprefixes them, so a writer asks doc.make["footer"]
whichever serialisation it was handed and never
doc.metadata["make.footer"].
3. The JSON columns were read by a character scan that stripped
brackets and quotes, which is wrong the moment a value holds an
escape: JSONValue.toString escapes "/" as "\/", so every heading
containing a slash came back with a backslash in it. All 36
document-languages differed on heading_ancestors_text. Now parsed as
JSON.
4. translation.ocn_aligned is in the database and not in the .ssp,
legitimately: it says whether the languages share one numbering, and
a .ssp is one language and cannot know. It is a fact about the file
rather than the document, so the loader leaves it out rather than
letting a writer tell which serialisation it was given.
Metadata values are also stripped now, as the text loader strips them:
one topic_register value carried surrounding spaces in the database
that the text form cannot represent.
After all four: 36 of 36 document-languages give byte identical dumps
from their .ssp and from their .ocda.db.
(assisted by Claude-Code)
|
|
The readers each read one serialisation and print. A writer needs
neither: it needs the document. So the model and both loaders live
here, and a writer cannot tell which serialisation it was handed,
exactly as spine own writers cannot.
ocda.model imports nothing, which is the property that makes writers
possible: they depend on the model and on neither loader. The model
and the .ssp loader are the ssp-reader code, moved and renamed to the
level they belong to: the model is neither the .ssp nor the .ocda.db.
ocda.load_db is new, and returns the same value from sqlite that
load_ssp returns from text. ocda.load is the one place that decides
what a path is.
The claim is only true if the two loaders agree, so test/agree.d asks:
load a document both ways and compare the models object by object.
It found a real defect on its first run. Eight of thirty-six
disagreed, and the cause was the .ssp loader, not the formats: an
object whose first text line is blank lost it, the loader having
decided whether to start a new line by asking whether the accumulated
text was empty, which is a different question from whether a line has
been seen. Seven documents of the sample set have verse opening on a
blank line and every one of them was wrong.
Nothing else had caught it: the two .ssp readers were compared on
section counts, which the bug does not disturb. It took loading one
document two ways and comparing the objects. Fixed here, and in
ssp-reader/dlang which carried the same code. 36 of 36 now agree.
(assisted by Claude-Code)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|