Re: [Gnu-arch-users] [BUG] tla cat-archive-log downloads .listing files unnecessarily

[Aaron Bentley]

Matthieu Moy wrote:
> Aaron Bentley <address@hidden> writes:
>
>
> > Matthieu Moy wrote:
> >
> > > Aaron Bentley <address@hidden> writes:
> > >
> > >
> > > > What version of tla is this?  It looks like earlier than 1.2.1.  1.2.1
> > > > and later shoul be a bit better than this.,
> > >
> > > Yes, that's still a 1.2. :-(
>
>
> 1.2.2rc1 performs better, but still isn't optimal.
>
> write(3, "GET /arch/xtla/.listing HTTP/1.1\r\nHost: 
> mani.sourceforge.jp\r\nUser-Agent: ar"..., 134) = 134
> write(3, "GET /arch/xtla/.listing HTTP/1.1\r\nHost: 
> mani.sourceforge.jp\r\nUser-Agent: ar"..., 134) = 134
> write(3, "GET /arch/xtla/=meta-info/mirror HTTP/1.1\r\nHost: 
> mani.sourceforge.jp\r\nUser-"..., 143) = 143
> write(3, "GET /arch/xtla/=meta-info/name HTTP/1.1\r\nHost: 
> mani.sourceforge.jp\r\nUser-Ag"..., 141) = 141
> write(3, "GET /arch/xtla/=meta-info/http-blows HTTP/1.1\r\nHost: 
> mani.sourceforge.jp\r\nU"..., 147) = 147
> write(3, "GET /arch/xtla/=meta-info/signed-archive HTTP/1.1\r\nHost: 
> mani.sourceforge.j"..., 151) = 151
> write(3, "GET /arch/xtla/xtla/xtla--jet/xtla--jet--0.2/patch-100/.listing 
> HTTP/1.1\r\nH"..., 174) = 174
> write(3, "GET /arch/xtla/xtla/xtla--jet/xtla--jet--0.2/patch-100/checksum 
> HTTP/1.1\r\nH"..., 174) = 174
> write(3, "GET /arch/xtla/xtla/xtla--jet/xtla--jet--0.2/patch-100/checksum.cacherev 
> HT"..., 183) = 183
> write(3, "GET /arch/xtla/xtla/xtla--jet/xtla--jet--0.2/patch-100/log 
> HTTP/1.1\r\nHost: "..., 169) = 169
>
> I don't know whether this is an artefact of strace, but it seems
> /arch/xtla/.listing is downloaded twice, I don't think
> checksum.cacherev is necessary, and patch-100/.listing could probably
> be downloaded only in case of diagnosis necessity. Not sure
> =meta-info/mirror is necessary either.

=meta-info/mirror and =meta-info/name are part of the implementation of 
a struct arch_archive in tla.  Lazy instantiation would require a change 
in the api used to access archives, and all the code that uses that api. 
 I don't know about the second xtla.listing-- that's probably not 
necessary.

If we made the revision existence test part of the libarch/archive.h 
interface, we could get rid of the cacherev checksum check.

Aaron

--
Aaron Bentley
Director of Technology
Panometrics, Inc.