Re: gcc_latest_snapshot CVS tag (was: gcc-ss-20030526 is now available)

[Kelley Cook <kcook34 at ford dot com>]

Joseph S. Meyers wrote:
> On Tue, 27 May 2003, Kelley Cook wrote:
>
>> A "CVS -d snapshot_date" will accomplish the same thing, if anybody
>> had
>> a reason to get a specific snapshot.
>
>> It will not; snapshots are not generated from CVS as of a fixed time
>> of day.

Then simply either always create the snapshots based on a UTC Midnight 
checkout using the -D option then or just save the UTC time that the 
snapshot is created into a history file.

Either way, if in the highly unlikely event that someone with CVS access 
actually does want to duplicate a snapshot with CVS then they'll time 
with the "-D" option.  This is a simple process detail.

To answer the -r -D problem that Geoff Keating mentioned, then you would 
need to follow the process at http://gcc.gnu.org/bugs/reghunt.html#get 
for CVS Branch

All of the above is also, in my opinion, overkill for the reasons given 
below.

> The principle of having those tags (absent generating snapshots from
> CVS checked out at a fixed time) is that when we have published
> something (in this case, a particular set of files as a snapshot with
> a given name) we should always be able to reproduce what we have 
published
> with that name at that time.  For future snapshots we could consider
> checking out with a given date and time specified (and including the 
date,
> time and branch in the snapshot announcement message), if this will work
> properly on a branch, to save on tags for future snapshots.

The problem with this statement is that the weekly Snapshots are 
works-in-progress with no guarantee that they will function properly.  
Although people might of made bug reports against specific snapshots 
early in the egcs days, I would think that almost nobody does now.  
However if they do, the first thing that the bug tester will check is if 
the bug still exists with either HEAD and the top of the current release 
branch.  If it doesn't then they close the bug.  In other words no-one 
cares if a specific snapshot has a transient bug that is already been fixed.

As strong evidence for this view, a perusal of the other GNU projects 
hosted on savavanah.gnu.org and sources.redhat.com do not have snapshot 
tags in their repositories.

BTW, This is not an really an argument for the sake of arguing.  There 
is a big usability downside to too many tags.  Go to the web site 
repository (example: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/), 
scroll to the bottom and try to search for a specific branch (like 
gcc-3_3-branch).  There are hundreds of entries in there, 90%+ of which 
have never and will never have been used.

Kelley Cook