Refresh button now always updates the commit to the latest SHA instead
of preserving the stale one in edit mode. Both create and update routes
verify the commit still exists on GitHub before persisting.
Mongoose treats `gist` as a nested path, not a sub-schema, so
set("gist", payload) mis-casts the inner subdoc arrays and fails
validation with 'Cast to [string] failed' at gist.files.0. Set each
subpath individually so the files/comments arrays cast correctly.
Surface the path/repo/url tied to an AnonymousError when it gets
serialized for logging — previously logs only carried name, message,
and httpStatus, which made file_not_found entries impossible to trace
back to a specific file or repo. Extract the existing detail formatting
out of toString() into a public detail() method, harden it against
AnonymizedFile getters that can throw, and have serializeError include
the result as a "detail" field.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Multi-stage Dockerfile with BuildKit npm cache mounts and a separate
prod-deps stage so source edits don't reinstall or prune.
- Tighter .dockerignore to shrink build context.
- Healthchecks: add start_period and tighten interval/retries so
containers report healthy as soon as the process is actually ready
instead of after a full polling interval.
- Move recoverStuckPreparing() off the startup critical path; the
recovery sweep now runs in the background after app.listen.
- depends_on uses condition: service_healthy and the obsolete
compose 'version' key is gone.
- New scripts/build.sh + scripts/deploy.sh: deploy.sh builds, exits
early if the image is unchanged, runs a blue/green streamer swap
(scale to 2N, wait healthy in parallel, drop olds), then recreates
the API with --no-deps to avoid compose's depends_on re-poll.
The LFS-pointer probe buffered up to 150 bytes before deciding whether
to forward the blob or swap to the raw URL. For blobs that fit entirely
in the probe, decide() ran from the source's end event and attached
data/end listeners to an already-ended stream, so out.end() was never
called. The response hung until upstream timed out and storage.write
left an incomplete cached copy, which then forced a re-fetch on every
subsequent read.
Pass a sourceEnded flag through decide() and end the output directly
when the source has already finished. Also skip the GitHub blob fetch
when the tree size is already over MAX_FILE_SIZE, surfacing
file_too_big instead of a translated 422.
Replace check-then-insert with atomic findOneAndUpdate upsert keyed on
externalId, plus a single E11000 retry fallback. Eliminates the duplicate
key race when two requests resolve the same gh_<id> concurrently.
* chore(deps): bump uuid and bullmq
Removes [uuid](https://github.com/uuidjs/uuid). It's no longer used after updating ancestor dependency [bullmq](https://github.com/taskforcesh/bullmq). These dependencies need to be updated together.
Removes `uuid`
Updates `bullmq` from 2.4.0 to 5.76.5
- [Release notes](https://github.com/taskforcesh/bullmq/releases)
- [Commits](https://github.com/taskforcesh/bullmq/compare/v2.4.0...v5.76.5)
---
updated-dependencies:
- dependency-name: bullmq
dependency-version: 5.76.5
dependency-type: direct:production
- dependency-name: uuid
dependency-version:
dependency-type: indirect
...
Signed-off-by: dependabot[bot] <support@github.com>
* fix(bullmq): adapt isRunning + getJobs typing for v5 API
Worker.isRunning became a method (was a property in v2), and
Queue.getJobs now requires a mutable JobType[] (was string[]).
* clean up
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: tdurieux <durieuxthomas@hotmail.com>
Co-authored-by: Thomas Durieux <5577568+tdurieux@users.noreply.github.com>
Files like .jsonl that mime-types doesn't know fell through to
application/octet-stream and rendered as "Unsupported binary file" in
the viewer. Replace istextorbinary with isbinaryfile for content-based
detection, and use mime-types for name-based classification with a
textual application/* allowlist.
The streaming transformer now defers classification when the name is
inconclusive and sniffs the first chunk before emitting "transform",
so route.ts and AnonymizedFile.ts get a content-aware Content-Type.
Whitelists .jsonl and .ndjson to short-circuit dataset files.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Follow-up review pass after the cache fixes turned up several bugs in
the same family — silent failures that look like success to the client,
plus content-correctness issues in the ZIP and per-file delivery paths.
- zipStream: stop calling archive.finalize() on upstream/parser errors.
That produced a valid-looking ZIP (200 OK, archive opens) silently
missing entries — same class as #694, but worse because the user has
no signal anything went wrong. Destroy the response on failure
instead so the client sees a connection drop.
- zipStream: apply per-repo image/pdf gates inside the entry handler.
The single-file /file/... endpoint refuses to serve those types
via AnonymizedFile.isFileSupported when image=false / pdf=false, but
the ZIP shipped them anyway — privacy-relevant for maintainers who
toggle image=false to suppress identifying screenshots. Threaded
contentOptions through both ZIP entry points (direct and streamer).
- GitHubUtils.getToken: validate the OAuth token-refresh response
before persisting. On a non-2xx response or a body without a string
token, we used to overwrite the stored token with `undefined`, which
then propagated as `Authorization: token undefined` to every API
call — 401 even on public repos, with the config.GITHUB_TOKEN
fallback unreachable because the field was no longer falsy.
- AnonymizedFile.send (streamer branch): forward Content-Type from the
upstream streamer response. got.stream(...).pipe(res) carries body
bytes only, so the parent response had no Content-Type and browsers
guessed (text rendered as download, etc.). Also resolve on
res.on("finish") in addition to "close" — keep-alive sockets stay
open long after the response is delivered, delaying countView().
- Repository.updateIfNeeded: persist a renamed source.repositoryName
even when the commit hasn't changed. Previously the new value lived
in memory only and was overwritten on the next reload, so the
rename detection ran every request.
- Repository.anonymize: stop materialising a dummy {path:"",name:""}
FileModel for empty repos. That row collided with the special case
in AnonymizedFile.getFileInfo and surfaced in unfiltered listings.
- streamer/route POST /: reject filePath segments containing ".." or
empty parts. Defence in depth — the parent server validates against
FileModel before calling, but the streamer joins filePath straight
into the storage path, so any future caller forwarding an
unvalidated path could traverse out of the repo root.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Follow-up to the GitHubStream cache fixes. The same poisoned-cache
class existed in the GitHubDownload path and a few related spots:
- GitHubDownload.download: wipe pre-existing state before extracting
and write a .anon-complete marker only after a successful extract.
On error, rm the partial cache so a retry starts clean. getFileContent
and getFiles now gate on the marker instead of "any file/folder
exists," so a half-extracted tree can never be served as canonical.
- GitHubDownload.getFileContent: validate cached file size against the
upstream FileModel size (via the new AnonymizedFile.size()), same
guard as GitHubStream. getFiles filters the marker from the listing.
- FileSystem.listFiles: drop the bogus stats.ino.toString() as sha.
An inode isn't a content hash; anything comparing it to a Git blob
sha would silently disagree. Leave undefined.
- S3.write: remove the fire-and-forget data.on("error") -> this.rm(...).
Multipart Upload doesn't commit partial objects, so there was nothing
to clean up, and the handler raced retries and could delete a
previously-good object on a transient source-stream hiccup. The
size-validated read path recovers from any other undersized objects.
- GitHubStream.resolveLfsPointer: drop the post-decision early-return
in blobStream.on("error"). Currently redundant with the inner
listener, but removes the future-refactor footgun.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
A failed/interrupted GitHub fetch could leave a 0-byte or truncated
file in the local cache. Subsequent reads happily streamed the empty
content as the file's body — visible to users as an "Empty file" with
HTTP 200. Reproduced on artifact-70B6/Lethe/configs.py (#694).
- FileSystem.write: stream into a sibling .tmp and rename into place
only on finish. Stream errors discard the tmp and leave any prior
cached file untouched. Drop the utf-8 encoding that was silently
corrupting binary blobs.
- GitHubStream.getFileContentCache: accept an expected size and treat
cached.size < expected as a poisoned cache (truncated fetch) → rm
and re-fetch. cached.size >= expected is accepted, which keeps
Git LFS-resolved files (whose FileModel.size is the pointer size)
working.
- AnonymizedFile: expose size() and pass it through to the streamer
alongside sha so the cache check has the upstream size.
Existing poisoned entries self-heal on next access.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
When a GitHub repo is renamed and looked up by its new name, the lookup
by name misses but a record with the same externalId still exists,
causing E11000 on save. Fall back to a lookup by externalId before
creating a new document.
Fixes#500
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Files tracked by Git LFS used to come out as the pointer text:
version https://git-lfs.github.com/spec/v1
oid sha256:...
size ...
…because GitHub's blob API returns the pointer, not the resolved
content. Detect that prefix on the first ~150 bytes of the blob stream
and switch to a fresh fetch via the web raw URL
(github.com/<owner>/<repo>/raw/<commit>/<path>), which auto-redirects
to media.githubusercontent.com and resolves the LFS object — auth
header carries through. Non-LFS files are forwarded through the
existing pipeline unchanged.
Fixes#95.
When the anonymizer doesn't change a slice's text, the streamer used
to push Buffer.from(out, "utf8") — which loses any invalid-UTF-8 bytes
in the input (replaced by U+FFFD via StringDecoder). Files
mistakenly classified as text (binary blobs without a known extension,
text with stray non-UTF-8 bytes, BOMs) came out corrupted even though
nothing in the term list matched.
Track the raw chunk bytes alongside the decoded `pending`. On flush —
where we have every byte buffered — emit the original buffer directly
when the output equals the input, so a pure passthrough is bit-exact.
In the streaming OVERLAP path, do the same when the decode for that
slice round-trips losslessly; fall back to encoded output otherwise
(unchanged from before for that case).
Also add the "missing_content" locale entry for the
/api/anonymize-preview route.
istextorbinary returns null for filenames with no extension, and the
isTextFile() guard treated null as "not text" — so terms in LICENSE,
COPYING, AUTHORS, README (extensionless), CHANGELOG, NOTICE, and
similar conventional filenames went through the binary passthrough
in AnonymizeTransformer and were never anonymized.
Add a small whitelist of these names ahead of the istextorbinary call.
Fixes#493.
When a user renamed the original GitHub repository, anonymous_github
kept calling oct.repos.get({owner, repo}) with the cached old name and
got 404, marking the link broken even though the repository still
existed at a new path.
Recover the numeric GitHub id from the dbModel's externalId
("gh_<id>"). On a 404 from the name-based lookup, fall back to
GET /repositories/<id>, which returns the renamed repo. The caller
already updates source.repositoryName from r.full_name afterwards.
Fixes#409.
The form's live README/PR preview was running its own copy of
ContentAnonimizer in the browser. The two implementations had been
drifting — recent fixes for word boundaries (#175/#249), accent
matching (#280), custom replacements (#285), and the diacritic-stripped
variants only landed on the server. Reviewers saw one anonymization;
authors composing the form saw another.
Add POST /api/anonymize-preview that takes a snippet (or a batch) plus
the user's options and runs them through the same ContentAnonimizer
the file route uses. Replace the client-side anonymizeReadme() body
with a debounced call to that endpoint. The PR view's
anonymizePrContent() runs as a synchronous template expression, so it
now reads from a {original -> anonymized} cache that's refreshed in
the background whenever the PR details, terms, or options change.
Single-flight + debounce keep the form responsive; an in-flight
request is dropped on the next change.
A term entered as "Anonymous=>ABC" now scrubs "Anonymous" to "ABC"
instead of "XXXX-N". Lets users keep anonymized identifiers valid in
source code (no hyphen) and align tokens between paper text and repo.
Indexing for default-mask terms is unchanged: a list of
"Alpha=>AAA", "Beta" still produces XXXX-2 for Beta.
Fixes#285.
The server set Accept-Ranges: none on every file response. For text we
anonymize on the fly so byte ranges aren't meaningful, but binary
entries pass through unchanged — and the explicit "none" header makes
some browsers refuse to play <video>/<audio> elements that would
otherwise fall back to a full download. Newly uploaded MP4s under the
inline-preview threshold rendered as a blank progress bar (#538).
Only set Accept-Ranges: none for text entries; let binary entries omit
it so the standard fallback kicks in.
Fixes#538.
Without the path, two different files in the same repo (same sha, same
anonymization options) shared an ETag. If a browser ever sent the cached
ETag for one file while requesting another, the server would have
returned 304 against the wrong cache entry. Fold the path into the
ETag so each file has its own fingerprint.
Follow-up to b3c1030 (#439).
Files were being served with Cache-Control: max-age=18144000 (210 days)
keyed only on the upstream ?v=<sha>. Editing the term list left the
same URL serving stale anonymized bytes — visible to users in regular
tabs but not in incognito. The previous fix-by-incognito recipe in #439
is exactly this.
Switch to ETag-based revalidation that fingerprints both the upstream
sha and the saved anonymization options, with Cache-Control:
no-cache, must-revalidate. Browsers now revalidate on every request and
get a 304 when nothing has changed, or fresh content as soon as terms,
image/link/etc. options are updated.
Fixes#439.
When a user added "Davó" to the term list, "Davo" elsewhere in the
content was left untouched (and vice versa). Each term now also runs a
diacritic-insensitive pass: ASCII Latin letters expand to a class
covering common accented siblings, with Unicode-aware lookaround
boundaries so the trailing boundary still fires next to "ó" etc.
Pure helpers moved into src/core/term-matching so the test file can
import them instead of duplicating the logic.
Fixes#280.
The streaming zip pipeline was constructing AnonymizeTransformer first and
then assigning opt.filePath afterwards. AnonymizeTransformer determines
isText in its constructor from opt.filePath, so every entry was classified
as binary and passed through unchanged — the downloaded zip leaked the
original (un-anonymized) terms even though the web view scrubbed them.
Pass filePath via the constructor so isText is computed correctly.
Fixes#342, #349.
Wrapping every user term as `\b${term}\b` silently dropped matches when
the term started or ended with a non-word char (e.g. `@tdurieux`,
`@author .*`), because JS `\b` only fires at a word/non-word transition.
Replace with `withWordBoundaries()`, which only emits `\b` on the side
where the term has a word-char edge.
Fixes#175, #249.
tdurieux