Source: ksh93u+m
Severity: wishlist
User: reproducible-bui...@lists.alioth.debian.org
Usertags: randomness
Dear Maintainer,
I'm an occasional volunteer with the Reproducible Builds[1] project, and
noticed recently that the ksh93u+m Debian package failed build reproducibility
testing[2] on amd64.
The autoconfiguration script[3] probes the build host for the user process
limit CHILD_MAX[4] - and the resulting value is compiled-into the /bin/ksh93
binary. This means that the binary package varies (is non-reproducible) on
build hosts that have different process limits configured.
Some additional notes / context to help figuring out a suitable value:
* The CHILD_MAX setting is a limit to the number of (sub)processes that a
user is allowed to run simultaneously.
* The compiled-in ksh93 CHILD_MAX value is only read if the _runtime_ system
getconf call[5] returns a zero/negative value (unlimited?).
* A zero or negative CHILD_MAX value seems unacceptable because it could
result in an inifinite loop[6] in the jobs.c code.
* The default value[7] of 1024 was introduced in the AT&T-distributed source
some time between Y2007 and Y2011 - it may still be an acceptable value.
* At the time of writing, the most recent Debian amd64 binary package for
source version ksh93u+m/1.0.7-1 has a compiled-in CHILD_MAX of 31626.
Regards,
James
[1] - https://www.reproducible-builds.org
[2] -
https://tests.reproducible-builds.org/debian/rb-pkg/unstable/arm64/ksh93u+m.html
[3] -
https://sources.debian.org/src/ksh93u%2Bm/1.0.7-1/src/lib/libast/comp/conf.sh/
[4] -
https://sources.debian.org/src/ksh93u%2Bm/1.0.7-1/src/lib/libast/comp/conf.tab/#L67
Note: this file's header explains each character in the line's "CDLMUX".
[5] -
https://sources.debian.org/src/ksh93u%2Bm/1.0.7-1/src/cmd/ksh93/sh/init.c/#L1274
[6] -
https://sources.debian.org/src/ksh93u%2Bm/1.0.7-1/src/cmd/ksh93/sh/jobs.c/#L1832-L1833
[7] -
https://sources.debian.org/src/ksh93u%2Bm/1.0.7-1/src/cmd/ksh93/sh/init.c/#L153-L155
