IRC channel logs

<zimoun>hi!

<zimoun>civodul: thanks for fixing etc/time-travel-manifest.scm. I was trying to debug it since we had spoken about.

<civodul> https://ci.guix.gnu.org/jobset/time-travel is still failing for reasons that are unclear to me though

<civodul>needs more work!

<civodul>probably a build failure somewhere

<zimoun>yeah. One makes it hard is the lack of log.

<zimoun>Aside, since we know how to rebuild the package with time-bombs, we could copy the binary substitutes. No?

<civodul>… once we’ve rebuilt them, sure

<civodul>:-)

<civodul>i plan to work on that, but time flies!

<zimoun>I have rebuilt them.

<zimoun>Somehow, my question is the alternative: Are we waiting The Right Thing for managing time-bombs and have a way to rebuild packages containing time-bombs? Or could we rebuild them manually and just store the results for sharing?

<civodul>well there’s no Right Thing in sight, only tricks to make it easier to rebuild “in the past”

<civodul>we could try a manual rebuild too

<civodul>‘guix time-machine --commit=v1.3.0 -- build openmpi --no-grafts -v1’ succeeded in seconds on berlin

<civodul>ACTION almost disappointed

<civodul>actually it’s taking substitutes from itself

<civodul>so at least these binaries are available

<rekado>so far for most reproducibility questions our approach has been to sidestep the problem

<rekado>we constrain file names, hash inputs, restrict variations, etc

<rekado>is it an option to fake the time in the build container?

<zimoun>I have tried some stuff with ’faketime’… and I failed. I ended up using good ol’ hard system-side time reset.

<zimoun>I think civodul “idea” is to have a “way” to build using whatever time in the build container. Where “way” is currently not very clear. :-)

<civodul>rekado: no, there’s a “time” namespace, but it only affects the monotonic clock

<civodul>so we need a full-blown VM “living in the past” to offload to

<civodul>not too bad, but it’s quite heavyweight

IRC channel logs

2023-09-12.log