This is the mail archive of the
mailing list for the GCC project.
Re: Bootstrap broken
- From: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- To: pedwards at disaster dot jaj dot com
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Thu, 6 Dec 01 15:24:18 EST
- Subject: Re: Bootstrap broken
I said "automating more of this," not automating /all/ of it. :-)
It is mostly automated, except for the "commit" command.
The vast majority of changes are only to a few files, so this sort of
semi-automatic procedure seems most appropriate, with very little
likelihod of error.
That's why a good patch-committing script does a 'cvs diff' for
/another/ visual last-minute inspection, and waits for human approval
before actually calling 'cvs commit' with a generated (but
human-approved) command line.
You're forgetting a major point: you can't have any files being
commited (including ChangeLog) updated in the repository between the
last cvs update and the cvs commit. If you put too many steps between
those, you can never do the commit. Given how the cross-county nework
behaves a number of times, it can often take more than two
update/commit cycles to get a clean ChangeLog. Putting a cvs of the
entire tree in there will make it very unlikely that the process you
propose will ever converge.
I do a cvs diff *after* the commit to doublecheck that I got everything, but
this time somebody noticed it before the cvs diff finished because the
network was slow.