Source: bmap-tools Followup-For: Bug #1081336 X-Debbugs-Cc: jo...@debian.org Control: tags -1 ftbfs
Dear Maintainer (and including Josch as requested), I'm able to replicate this using a podman-based Debian trixie container (currently, using podman 5.3.1+ds1-7 on amd64, and running an amd64 container image of debian:trixie with ID a1b129fd683d) when building repeatedly with dpkg-buildpackage. So far I don't have much to add in terms of findings; all I have achieved is to seem to rule-out the possibility that random selection of the three variables for progress-bar, sync, and verify prepared[1] for each test copy operation are a factor -- because I've encountered stalled test_api_base behaviour regardless of the runtime true/false value for each of those variables, assuming they're independent. (_strictly_ speaking there could remain complex boolean interactions between them that are necessary the stall preconditions -- I can't disprove that because I haven't encountered all eight permutations possible in the truth table; but it seems likely to me that they are independent variables) PSV (P = progressbar, S = sync, V = verify) stall counts --- FFF 1 FTT 1 TFF 1 TFT 2 TTF 2 The stall encounter rate, across the total number of testsuite evaluations attempted, seemed to be around 15-20% on my local machine. Not much in the way of clues/indicators yet, but it leads me to believe that focusing elsewhere than the random variables (e.g. on the file and pipe operations) may be a good next step. Regards, James [1] - https://sources.debian.org/src/bmap-tools/3.8.0-1/tests/helpers.py/#L301-L304
