Re: Enhancing lintian performance for large source packages
Andrius Merkys <merkys@debian.org> writes:
> NYTProf highlighted that over 70 % of the runtime was spent in searching
> for longest lines in source files as part of
> Lintian::Check::Files::SourceMissing and
> Lintian::Check::Files::Contents::LineLength checks. I modified them to use
> better performing solutions for the same task and managed to decrease
> their time share to 32 % for the package in question. This reduced lintian
> runtime from 2278s to 820s for the package in question. I opened three MRs
> for lintian with these changes [2][3][4].
This is fantastic work. Thank you!
Given that the time share is still 32%, though, I'm going to make an
additional controversial suggestion: perhaps we should provide an easy
documented shorthand to tell Lintian to not run "expensive" checks, or
perhaps not those specific checks? I know the -X flag already exists, but
I think people don't know how to find it or determine what to pass to it.
(The Lintian man page says to see the CHECKS section, but then contains no
CHECKS section.)
Personally, I have never found either of those checks useful to me as a
packager. They're both a source of constant annoying false positives and
have never once alerted me to a problem that I didn't already know about.
I have considered, in the past, arguing that Lintian should remove them
entirely given how expensive they are and how prone to false positives
they are, but if there were some way that I could easily just opt out (and
of course assuming that ftp-master never rejects based on those checks),
that would relieve a lot of the pressure.
--
Russ Allbery (rra@debian.org) <https://www.eyrie.org/~eagle/>
Reply to: