marram · notifications since Nov 7

Seven notifications need your call

I handled 36 of the 43 notifications from your week away, and these seven need you. I post, merge or close nothing on them until you answer, and act on each answer as it comes in.

issue #423 · Rhea

Close the Windows config report as a duplicate?

marram ignores Rhea's ~/.config/marram/marram.toml on Windows. That's #397, which #402 fixed on main by reading %APPDATA%\marram; it ships in 1.7.3. Either way I post:

Thanks for the clear report! This is #397: from 1.7.3, marram reads %APPDATA%\marram\marram.toml on Windows. Until then, pass that path with --config.

Thanks for the clear report! This is #397: from 1.7.3, marram reads `%APPDATA%\marram\marram.toml` on Windows. Until then, pass that path with `--config`.

I'd close it: #397 tracks the fix, and the reply names the release.

Rhea hears at 1.7.3 Leave #423 open until 1.7.3 ships Rhea follows #397 Close #423 as a duplicate of #397

Dependabot · #421

Fix and merge the reqwest 0.13 update?

The bump from reqwest 0.12.24 to 0.13.1 fails to build, because 0.13 renamed the rustls-tls feature to rustls:

 [dependencies.reqwest]
-version = "0.12.24"
-features = ["rustls-tls"]
+version = "0.13.1"
+features = ["rustls"]
 default-features = false

I'd merge it. With the rename the suite passes on all three platforms. 0.13 also checks certificates against the operating system's store instead of a bundled list, and all 1,212 links in the HTTPS fixture resolve the same under it.

reopens at 0.13.2 Close #421 and stay on 0.12 1 line 1,212 links pass Push the rename to #421 and merge it

review requested · Tomasz's #419

Approve moving CI to cargo-nextest?

#419 runs the suite with cargo-nextest, which runs each test in its own process, in parallel.

CI per push
6 min 40 s → 3 min 10 s
Flaky test
the redirect-loop test failed 2 of 30 runs while another test held its port; #419 runs it serially, and 30 more runs passed
Doctests
nextest skips them, so #419 adds a cargo test --doc step

I'd approve it: every push gets its result three and a half minutes sooner, and running one test serially costs about a second. Giving each test its own port is the real fix for the flake, and it can follow on its own.

CI 6m 40s no workaround Ask Tomasz to give each test a free port first #419 waits for that change. CI 3m 10s 1 serial test Approve #419 I'll open an issue for the free ports.

discussion #415 · mentions you

Promise Ade's team a quarter for 2.0?

Ade's docs team is choosing a link checker for 30 repositories and asks whether 2.0 will break the 1.x config. I drafted the reply from ROADMAP.md:

2.0 is planned for the first quarter of next year. It reads 1.x config files unchanged and warns about each key that moved, and marram migrate-config rewrites a file in place. Adopting the 1.x format now is safe.

2.0 is planned for the first quarter of next year. It reads 1.x config files unchanged and warns about each key that moved, and `marram migrate-config` rewrites a file in place. Adopting the 1.x format now is safe.

I'd post it without the date. What Ade asked about, the config format, is settled; the roadmap marks the quarter tentative, and a team planning 30 repositories around it would hold you to it.

promises the format Post it without the date I'll drop the first sentence. promises Q1 too Post it as written

pull request #418 · Jaana's first contribution

Merge --stdin as it is?

#418 reads the files to check from standard input, one path per line, so git ls-files '*.md' | marram --stdin checks only tracked docs. Its tests pass on Linux, macOS and Windows, and it follows the existing argument handling.

One gap: without -z, git ls-files quotes any path holding a non-ASCII character or a newline, so docs/ja/入門.md arrives as "docs/ja/\345\205\245\351\226\200.md" and fails to open. git ls-files -z writes each path whole, ended by a NUL byte, the way xargs -0 reads them. I'd ask for it before merging: a translated docs tree hits this on its first run, and the change is a few lines. The review I'd post:

Thanks, this is a clean change and the tests cover it well. Before I merge, could you add -0 so --stdin can also read NUL-separated paths, as git ls-files -z writes them? Without it, git quotes a name like docs/ja/入門.md, which then doesn't open. Happy to point you at where the parser would take it.

Thanks, this is a clean change and the tests cover it well. Before I merge, could you add `-0` so `--stdin` can also read NUL-separated paths, as `git ls-files -z` writes them? Without it, git quotes a name like `docs/ja/入門.md`, which then doesn't open. Happy to point you at where the parser would take it.
1 more round all paths open Post the review asking for -0 merged today ASCII paths only Approve and merge #418 I'll open an issue for NUL-separated input.

issue #424 · 14 upvotes

Where does anchor checking go?

The most-upvoted open request: report a link whose #fragment names no heading or id on the page it points to. Each target's HTML has to be parsed, which made a full check of marram's own docs 40% slower in my prototype, so it would sit behind --anchors.

I'd put it on the 1.8 milestone: the prototype already works, and the slowdown reaches only those who pass the flag. Building it takes three days, after 1.7.3.

0 days never ships Decline it as out of scope Close #424 and point to anchor checking in a site generator's build. 0 days if contributed Label it help wanted With a note on where the parser would hook in. 3 days ships in 1.8 Put it on the 1.8 milestone

private security report · Nadia, Nov 9

Open an advisory for redirects into internal networks?

marram follows redirects wherever they lead, so a docs link that redirects to 169.254.169.254 makes CI request the cloud metadata service and log its status. The body never reaches the log: an attacker learns which internal hosts answer, not what they hold.

Reproduced
1.7.2 on a self-hosted EC2 runner logged 200 OK
Nadia asks
a CVE, and public disclosure on Dec 9
Fix
refuse redirects from a public address into a private or link-local one unless --allow-private is set

I'd open the advisory: the fix is small, and 1.7.3 ships on Nov 27, before Nadia's date. A team whose public links redirect into its own network, such as a sign-on page, would need the flag.

0 days still probeable Reply that this is expected behavior Document that marram follows redirects, and leave CI runners to restrict their network. 1 day SSO needs flag Open a private advisory and fix it in 1.7.3 Request a CVE, credit Nadia, and publish the advisory with the release.
The 36 I handled without you