On Wed, Dec 09, 2015 at 09:56:37PM +0100, Ludovic Courtès wrote: > Leo Famulari <[email protected]> skribis: > > > On Tue, Dec 08, 2015 at 09:52:30AM +0100, Ludovic Courtès wrote: > >> Mark H Weaver <[email protected]> skribis: > >> > >> > Ricardo Wurmus <[email protected]> writes: > >> > > >> >>> +(define-public swh-plugins-lv2 > >> >>> + (let ((commit "5098e09e255eaed14e0d40ca5e7e6dfcb782d7ea")) > >> >> > >> >> We usually don’t use full commit hashes. You could probably trim it to > >> >> the first six characters or so. > >> > > >> > I would recommend using at least 10 characters, maybe more. We should > >> > use enough characters to ensure that the commit id remains unique for as > >> > long as this package version remains in use -- keeping in mind that for > >> > purposes of reproducing old experiments, someone might try to build this > >> > package+version several years from now. > >> > >> I sympathize with this. I would think 10 digits is more than needed, > >> though: With 6 hex digits, it takes on average 16^6 = 16M commits before > >> the 6-digit ID is ambiguous, and with 8 hex digits that goes to > >> 4 billion commits (Emacs has around 123,000 commits as an example.) > >> > >> But anyway, to be super-safe, we could use the full SHA1 in the URL, but > >> strip it in the ‘version’ field so that it remains readable. > >> > >> Thoughts? > > > > I don't see a reason to not use the full hash. I know the odds but why > > discard the information? We can't complain about it being ugly given the > > aesthetics of /gnu/store. ;) > > The ‘version’ field shows up in user interfaces, where aesthetics do > matter IMO. WDYT? > > (In some cases, it’s also a good way to avoid hitting limitations on the > shebang length and on the AF_UNIX socket name length.)
Okay, then I like your idea of using the full SHA1 in URLs but truncating it in the 'version' field. > > Ludo’.
