racket / pkg-build Goto Github PK

Packages using a specific older version for a dependency via a valid package source format that relies on a web resource (i.e. hosted Git repo) fail to build on the package server, even though the packages themselves are in conformity with prescribed package management concepts.

The error reported by the package server is:

Querying Git references for qi-lib at git://github.com/countvajhula/qi.git?path=qi-lib#v1.0-maintenance
pkg: Contacting github.com
pkg: Network error; retrying after 0.1s
pkg: Contacting github.com
pkg: Network error; retrying after 0.2s
pkg: Contacting github.com
pkg: Network error; retrying after 0.4s
pkg: Contacting github.com
pkg: Network error; retrying after 0.8s
pkg: Contacting github.com
pkg: Network error; retrying after 1.6s
pkg: Contacting github.com
tcp-connect: host not found
  hostname: github.com
  port number: 443
  system error: Temporary failure in name resolution; gai_err=-3
  context...:
   /home/root/racket/collects/openssl/mzssl.rkt:1935:0: do-ssl-connect
   /home/root/racket/collects/racket/contract/private/arrow-val-first.rkt:555:3
   /home/root/racket/collects/net/http-client.rkt:67:0: http-conn-open!
   /home/root/racket/collects/net/http-client.rkt:274:0: http-conn-open
   /home/root/racket/collects/racket/contract/private/arrow-val-first.rkt:555:3
   /home/root/racket/collects/net/url.rkt:202:0: http://getpost-impure-port
   /home/root/racket/collects/net/url.rkt:305:0: get-pure-port/headers
   /home/root/racket/collects/racket/contract/private/arrow-val-first.rkt:555:3
   /home/root/racket/collects/net/git-checkout.rkt:245:0: initial-connect
   /home/root/racket/collects/net/git-checkout.rkt:55:2: retry-loop
   /home/root/racket/collects/pkg/private/stage.rkt:62:2: lookup-normally
   /home/root/racket/collects/pkg/private/stage.rkt:115:0: stage-package/info
   /home/root/racket/collects/pkg/private/install.rkt:141:0: install-packages
   /home/root/racket/collects/pkg/private/install.rkt:925:4
   /home/root/racket/collects/racket/contract/private/arrow-val-first.rkt:555:3
   /home/root/racket/collects/racket/file.rkt:763:8
   ...
The time is now Friday, June 17th, 2022 12:37:17am

This means that packages that use an officially supported dependency mechanism are incorrectly reported as broken, causing docs and all dependent packages to also be reported as broken.

For some context, the relation package is faced with the choice of either continuing to support legacy Racket code (7.8+) or drop support for <8.3. It could do the former by using a Git ref to a maintenance branch of one of its dependencies (Qi), but currently this runs into the present issue. The other option is to drop support for <8.3 which sweeps it under the rug, but seems a bit extreme.

Package affected: unlike-assets as of Feb 16th. It had that status most of that day, if not all day.

The status links are by the bullet points below. I expect these links won't always resolve to the same document, so note that the failure link 404'd at this time of writing. However, it did earlier resolve to a very short error message that I do not recall (with apologies).

I'm assuming that this build actually succeeded since the 404 only applies to the failure log now.

• https://pkg-build.racket-lang.org/server/built/fail/unlike-assets.txt
• https://pkg-build.racket-lang.org/server/built/install/unlike-assets.txt

Sometimes, we expect certain tests to fail specifically when they are run in the build environment.

For example, racket-packet-socket tries to get raw access to an ethernet device, which will clearly fail in the build sandbox. This makes it look like there are failing tests in racket-packet-socket in all the places that the build-server's output is shown to users (e.g. build-server UI itself; package catalog UI), even though that's sort-of not really the case.

It would be nice to have some way of saying either "I expect this and that file to fail its tests in this or that way when run on the build server", or "O build-server, do not run tests for this and that file!"

The rhombus-prototype build failed a few times by being killed. I think it’s running up against a 1GB memory limit.

Of course, 1GB is a lot… Part of the issue is that pkg-build uses -j 2, so a package can get past 1GB by having two pieces that are each in the neighborhood of 500MB. That amount of memory use for one document is not unprecedented, as illustrated by peaks in the blue and green regions of https://build-plot-cs.racket-lang.org/. That is, the main-distribution build stays under 1GB only with -j 1.

I will continue efforts to make rhombus-prototype build in less memory, but even if I succeed, the issue is likely to show up again with some package. Would it be possible and make sense to increase the memory limit, either to 1.5GB ot 2GB? Or could pkg-build provide a way for a package to request -j 1?