On Tuesday, February 21, 2017 9:51:06 AM CET Daniel Kahn Gillmor wrote: > On Tue 2017-02-21 06:31:57 -0500, Tim Ruehsen wrote: > > both, /usr/share/publicsuffix/public_suffix_list.dafsa and > > /usr/share/publicsuffix/public_suffix_list.dat have the same file date > > (mtime). > > > > Together with a small glitch in libpsl/psl_latest() this leads to .dat > > being loaded which causes lot's of parse/conversion/malloc overhead. > > whoops, thanks for catching this! > > > There are two possible solutions: > > 1. The .dafsa file should be older than the .dat file by at least 1s. > > This makes sense since the .dafsa file is generated from .dat and thus > > should be older. > > as you said in your followup, the .dafsa should be *newer* than the > .dat. I'm trying to decide how to package this change most cleanly. > > The timestamps for the installed files are "clamped" to the timestamp in > the debian/changelog entry. This is a relatively new change, in debian, > the result of the reproducible-builds policy, which is the right one. > > Since the build daemons are all going to do the build *after* the > debian/changelog entry timestamp, just including a "sleep 1" before > invoking psl-make-dafsa won't provide a fix. > > > Also the .dat file should have the original timestamp from upstream. > > What timestamp do you mean by this? I don't think that the upstream > development includes a specific timestamp aside from those found in the > git commit history. Do you mean that the .dat should have timestamp > based on the git commit timestamp? if so, which timestamp -- the author > timestamp or the committer timestamp? Given that i'm naming the > upstream package version number based on the committer timestamp, that > might be a reasonable way to choose the timestamp for the .dat file. > > if we do that, then should we let the .dafsa receive the clamped timestamp > of the debian/changelog? or should we explicitly set the .dafsa to the > .dat + 1 second?
Technically it doesn't matter as long as the .dafsa is at least 1s newer than the .dat. The timestamp of the latest commit/change would be fine, though. Maybe this can be used by 'date' and/or 'touch' ? git log -n 1 --pretty=format:%cd --date=iso public_suffix_list.dat > > 2. Apply a 1-line patch to libpsl (already in upstream git repo) which > > prefers the distribution (.dafsa) file. > > I think this is worth applying anyway, since it's the right thing to > do. thanks! I've cloned this bug report and reassigned the clone to > libpsl to track that.
signature.asc
Description: This is a digitally signed message part.
