Repository navigation
is_noop calls a :ref: with display text markup, so its prose is never translated #54
Description
Activity
Measured against the pinned mirror, so the size is not a guess.
139
noopentries carry role display text of two or more English words, 630 words in total. Every one is prose a reader sees, and none of them has ever been queued or checked.That is the lower bound. It counts two or more alphabetic words to keep single tokens out, since a one-word display text is often an identifier and would need the same judgement
is_version_markermakes.Shipped in #59, content tamnd/python-docs-vi-machine-translation#55.
130 entries rather than the 139 this issue measured. The difference is how the words are counted, and the recount is the part worth keeping.
Counting runs of
[A-Za-z]{2,}over the display text finds 266 entries, and 136 of those are wrong.:c:member:base_exec_prefix <PyConfig.base_exec_prefix>`` is one identifier, and a pattern looking for runs of letters readsbase, `exec` and `prefix` in it and calls that a sentence. Counting whole words between the spaces gives 130.That leaves 81
:c:member:entries with a one-word display text as no-ops, which is the same callis_version_markermakes about a bare token and deliberately the cautious one. A missed entry stays exactly as it is today. A false positive sends an identifier to a model to be translated into something that links nowhere.Where the 130 live:
role entries :ref:117 :rfc:6 :term:4 :doc:3 :option:2 All 130 display strings were read rather than sampled, and there is no identifier among them:
A logging cookbook,Basic Tutorial,Configurable memory allocators,Explanation of the object-oriented interface,Command Line Interface Libraries.prose 75,593 to 75,723, noop 5,721 to 5,591, batches 2,801 to 2,803. The audit does not move, which is the expected result: these entries were invisible to every check and are now untranslated, and an untranslated entry is not a finding. They will be queued on the next fleet run.
is_noopstrips the markup and asks whether anything is left. For a role with display text there is prose inside the markup, and stripping takes it away with everything else.Found while narrowing
human_segmentsto skip code. One entry in the mirror is:A person translated it correctly, rendering the display text and leaving the target alone:
The classifier calls it
noop, which means two things.batchnever queues it, so no model is ever asked. AndCorpus.prose()skips it, so no check that reads a translation looks at it. It is only visible at all because it happens to be the one non-prose entry a human translated to something other than a copy of the source.This is the failure the
is_version_markerdocstring describes, in a different rule: an entry that wants translating, classified as one that does not, invisible from every angle that would report it.:ref:targetwith no display text is a genuine no-op and should stay one. The distinction is whether the role carries text a reader sees, which istext <target>`` inside the backticks, andP03and `P06` already parse exactly that shape.Not urgent on its own, one entry. Worth measuring first: the count of
noopentries whose markup contains display text is the real size of this, and it is a lower bound on entries the pipeline will never translate and never report.