This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: removing old snapshot CVS tags? (was: gcc_latest_snapshot CVStag)

On Sun, 13 Jul 2003, Gerald Pfeifer wrote:

> And if so, they will use the mailing list archives and hardly use CVS
> tags (not the least because machines operational in a few years will
> not be able to bootstrap such old versions of EGCS and GCC).

Naturally if building the code they'd use an appropriate contemporaneous
environment (possibly under emulation).  I'm actually supposing that it
would be more a matter of referring to particular code versions in support
of studies starting through mailing list archives.  I am well aware that
some studies have found contextual material of more use for historical
research than the actual software (e.g. Martin Campbell-Kelly, "Software
preservation: Accumulation and simulation", IEEE Annals of the History of
Computing, vol 24, no 1, pp 95-96); it's from consideration of such
matters that I think _given_ the rich contextual material available in the
form of mailing list archives there will then be value in being able to
refer to the source of a few particular versions determined from the study
of the list archives etc..

> Frankly, I think this is a strawman argument, and one for which we can
> hardly afford 1GB out of 3GB of free disk in ~ftp/pub/gcc.

A tar.bz2 of the repository does not occupy 1GB!  (That's more like the
full uncompressed repository.)  The space saved in the repository by
removing tags would be insignificant and I haven't seen any evidence of
measurable speed cost of tags existing in files either.  And of course the
information content of the tags can be reproduced in a form much more
compact than a tarball of the whole repository, if both 100MB and the tags
being in the repository are still concerns, e.g. a list of cvs commands
assigning the tags to the correct versions in every file (and tested on an
rsynced copy to restore the repository state correctly).

Joseph S. Myers

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]