Skip to content

Integrate nine robustness fixes for 1.6.0 (#107 #208 #209 #210 #211 #231 #234 #236 #240) - #252

Merged
danielplohmann merged 28 commits into
masterfrom
integrate/robust-B
Sep 30, 2026
Merged

danielplohmann merged 28 commits into
masterfrom
integrate/robust-B

Conversation

@danielplohmann

Copy link
Copy Markdown
Collaborator

Nine robustness fixes for 1.6.0, integrated the same way as #248 and #251. Each PR's head commit is an ancestor of this branch, so GitHub marks them merged when this lands with a merge commit.

PR fixes
#107 an empty result, a failed job and a terminated job each get their own page, instead of "job not found"
#208 malformed function-ID lists reach the backend instead of raising an unbound-name error
#209 conditional icon tags are closed in both branches
#210 ADR references point at the decisions they describe
#211 submitting a binary no longer leaves a SHA-256-named copy on the web host, which also removes a path derived from user input
#231 the username/password handlers accept only POST, and the "username taken" check runs only after the password is proven
#234 link hunt defaults the family count when the form leaves it empty
#236 the unnamed family's samples (family id 0) are listed in a cross compare's job row
#240 a result whose job type the dispatch doesn't know renders a page instead of a 500

Resolutions

Verified

Not in this batch

🤖 Generated with Claude Code

https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8

r0ny123 and others added 28 commits August 29, 2026 20:14
data.result opens with `if result_json:`, and an empty dict is falsy, so a job
that finished and produced an empty report skipped the dispatch entirely, failed
the "not finished" guard because it had finished, and landed in the else that
renders result_invalid.html - "Job ID ... or the result referenced by this job
was not found in the system." The reader is told their job does not exist.

The tail is now four explicit cases - unknown id, failed or terminated, finished
with an empty report, still running - which is also what makes the empty case
sayable at all. result_empty.html already existed and already says it.

Failed and terminated are checked before finished so a job that is both keeps
the answer it had.

While in here: the dispatch chain had no else either, so a job method it does
not know returned None and Flask answered 500. That is the same crash as the one
just fixed in data.linkhunt; nothing reaches it today because the chain covers
every method mcrit currently defines, but a new backend method would.
Every unit job on every Python version fails with:

    /opt/hostedtoolcache/Python/3.11.16/x64/bin/python: No module named pytest

The Makefile said "pytest and pytest-cov arrive with mcrit, so requirements.txt
covers both". mcrit declares them under its `dev` extra:

    $ python -c "import importlib.metadata as m; print(m.requires('mcrit'))"
    ... 'pytest; extra == "dev"', 'pytest-cov; extra == "dev"' ...

so a plain `pip install -r requirements.txt` does not bring them in, and the
suite has not actually run in CI since mcrit moved them there. Ruff is green
throughout, which is why this looked like a working pipeline.

Installed in the workflow rather than added to requirements.txt: neither is a
runtime dependency and a deployment should not get them. Pinned, for the same
reason the lint job pins ruff. The two stale comments that said otherwise are
corrected.

(cherry picked from commit 8268666)
work/harness/cookies.txt is a cookie jar written by a local debugging
harness - a Flask session signed with a secret hardcoded in that harness,
against a throwaway SQLite database. Not a credential for anything, but it
has no business in this change. It arrived with the cherry-picked CI fix.
Review found this PR fixed empty-vs-unknown and then merged three further
reasons back together.

  * A failed or terminated job rendered result_invalid.html, which says the
    job "was not found in the system". That is false in both halves: the job
    is in the system and the view is holding its Job object. Reporting one
    reason as another is the thing the issue is about, so doing it in the
    fix was the wrong shape. job_failed.html now says what happened, and
    carries the job's own detail table.

  * getResultForJob answers None both when the job recorded no result and
    when the document behind that id can no longer be fetched - a purged
    GridFS entry, a backend re-provisioned with the job metadata intact. The
    discriminator, job_info.result, was sitting right there unconsulted, so a
    report that merely cannot be retrieved was announced as "Job ID X does
    not contain any data". In a triage UI that reads as "your matching run
    found nothing", which is a wrong analytical answer rather than wrong
    wording. result_invalid.html already says the accurate thing - the result
    referenced by this job was not found.

Telling an empty report from an absent one needs the fetch result kept apart
from the cached one, since {} and None are both falsy; `fetched` does that,
and the comment says why it cannot be folded back in.

Also dropped a comment that described an impossible state: is_failed is
attempts_left == 0, decremented only in Job.error(), which never sets
finished_at, so no job is ever both failed and finished.

NOTE: PR #2 needs the same job_failed.html for linkhunt's dispatch and adds
an identical copy, so whichever merges second will conflict add/add on that
one file; either copy is correct.
`init` still ran `pip install -r requirements.txt` alone, so a fresh checkout
that follows the README lands without pytest and `make test` fails with
"No module named pytest" - the same failure the CI change on this branch fixes
for the runner, left unfixed for a contributor.

pytest and pytest-cov are not runtime dependencies and are not in
requirements.txt; mcrit declares them under its `dev` extra, which nothing
here requests, so they do not arrive transitively either. Install them from
`init`, pinned to the versions CI uses, and say so once above the target
rather than in a comment further down that describes a step that never ran.

Reported by Codex review.
Conflicts in tests/testResultPages.py only. Imports are unions. The job
page test takes master's parametrized form, after this branch's #73 tests.

One collision the conflict markers did not show: master's linkhunt work
(1f38998) added test_a_terminated_job_says_it_was_terminated for
/data/linkhunt, and this branch has a test of that name for /data/result.
Python keeps the later definition, so master's would have stopped running
without an error. Renamed this branch's to name the page it covers; ruff's
F811 is what caught it.

991 passed, ruff clean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011EAW1DRkBwmjZtGzQ5pgDA
Compute the API label flag before either POST branch so malformed ID lists can reach the backend's 400 response. Remove the unused sample search from hash-list cross comparison; its failure path referenced an unbound name and did not affect the redirect. Cover both paths with route tests.

Fixes #201
Name the actual ADR files in code comments and keep the linked numbers in AGENTS.md aligned. Add a small reference check so future renumbering cannot leave misleading links behind.
Submit forwarded the binary to the backend but also wrote a SHA-256 named copy under temp/uploads. No route reads that copy; query promotion reads files named by backend job ID. Remove the extra write and cover both mapped and unmapped submissions. Document which existing files can be removed.
Every module under tests/ starts with the shebang and the LOG/logging
boilerplate; this one did not.

The lint also indexed past "{% endif %}" without checking the line had
one, so an icon written with "{%- endif %}" or with its endif on the next
line would have stopped the test with an IndexError instead of failing on
the template line that needs a look.
Every module under tests/ starts with the shebang and the LOG/logging
boilerplate; this one did not.

Each of the three checks passes on an empty list of matches, so a pattern
that stopped matching - a moved docs tree, a new link style - would pass
without checking anything. The added test asserts that each pattern still
finds references in the searched tree.
The version history's pending entries are written " * unreleased:", as
in #160 and #212; this one was capitalised. Every module under tests/
starts with the shebang and the LOG/logging boilerplate, and this one
did not.
With the unbound name fixed, a body that is not an ID list gets the
backend's answer. But one that is could still be taken for malformed:
the branch matched `request.data`, which Flask leaves empty for a form
content type after parsing the body into `request.form`, and a form
content type is what `curl --data "1,2"` sends by default. mcrit reads
the raw request stream whatever the type, so the same call succeeded
against the backend and failed through the passthrough.

Read the body with `request.get_data()`, as the binary query branch in
this router already does. Measured against mcrit 1.9.0 with
`curl -X POST --data 1,2 .../api/functions`: master answers 500
(UnboundLocalError), the previous commit 400, this one 200 with the two
functions. A body that is not an ID list still gets the backend's 400.
change_username and change_password ran every check as an independent
if, so a malformed username still paid for check_password_hash and a
username lookup, and a mismatched confirmation still hashed the old
password. The checks now short-circuit. A username change checks the
format, then the password, then whether the name is taken - the
password deliberately ahead of that cheaper lookup, so that a caller
with the wrong password is no longer told whether a name exists.
Previously the last failing check won, and "already taken" was shown
even when the password was wrong. A password change checks the
confirmation, then the password.

All three account handlers, regenerate_apitoken included, re-read the
caller's row although load_logged_in_user has already read it by the
same session id for this request and login_required guarantees it is
there. They use g.user now, which also makes the ad-hoc session id
validation redundant.

Both form routes accepted GET while indexing request.form, so a GET
ended in a 400 (a 500 under FLASK_DEBUG). They are POST-only now, as
AGENTS.md asks of writing routes; the settings forms already post.

Fixes #206.
The link hunt's filter form sends every field, and an empty one parses
to None. Both presets set filter_unpenalized_family_count to 2, but a
filter the user filled in passed None on to mcrit's getLinkHuntResults,
which compares the count to an int: a TypeError, and a 500 for any
filter submitted with that field cleared.

An empty count now gets the presets' 2, and the form shows it.
job_description groups a cross compare's samples by family, and skipped
every sample whose family_id was falsy, which family 0 is. Every mcrit
storage keeps family 0 as the unnamed family, and a sample submitted
without a family lands in it, so the row of a cross compare over such
samples left all of them out, below a count that includes them. Every
other description in the template shows family 0 as "Unnamed", through
format_family_name, and so does this one now: the check lets through
anything but None, the case it was written for.
The dispatch chain in data.result() had no else branch, so a finished
job with a result but a method it does not list fell off the end and
returned None, which Flask answers with a 500. On a live mcrit 1.9.0
instance that is the result page of each of the three repair jobs
(repairMinHashes, recomputeFamilyStats, rebuildPicBlockHashIndex).

Answer such a job with result_incompatible.html instead, and cover it
with a test using a job method nothing knows.
… search

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
…ept only POST

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
… it empty

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
README conflict: both #208 and #211 add an "unreleased" line to the version history; kept both.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
…es not know

#107 already contains #240's whole code change (the else branch rendering result_incompatible.html
for an unknown job type), so data.py keeps #107's version; the tests keep both sides, so #240's
test checks that #107's branch covers its case.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
…r has

Since #228 the API passthrough reads Content-Type and the raw body; #208's tests,
written before that, failed on the fake lacking both. Test-only change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HERmVhHo4ibp7HSTXr3Ya8
@danielplohmann
danielplohmann merged commit 1cdc195 into master Sep 30, 2026
8 checks passed
@r0ny123
r0ny123 deleted the integrate/robust-B branch October 30, 2026 06:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants