Re: [Gnu-arch-users] GCC v. Arch address@hidden: Regressions on mainline]

[Tobias C. Rittweiler]

On Wednesday, June 23, 2004 at 5:04:59 PM, 
    Colin Walters <address@hidden> wrote:

> On Wed, 2004-06-23 at 14:36 +0200, Tobias C. Rittweiler wrote:
>
> > Don't test on *every* commit, but firstly pile those [commits] up. Just
> > from time to time do a build-test-cycle:
>
> That breaks the invariant that *every* revision on the mainline passes
> the testsuite.  Sometimes you might need to back up by changesets to
> narrow down a bug that *wasn't* found by the test suite.  Knowing that
> all the revisions there did currently pass is important.

Even though that's certainly a very true point, it's more or less
completely irrelevant to Tom's current thought-game (of course, only
AFAICS).

Tom just guessed that GCC *might* fall into this or a similiar case
of his fictive project which has two properties:

  1. Changes are happening lighning fast.

  2. Running the test suits takes a huge amount of time compared to the
     time new changes are hissing in.

For very large values of both properties, running the test suite for
each incoming changeset is way too expensive such that the queue of
incoming requests would expand drastically and it would steadily take
more and more time until the currently most recent requests are
processed by the PQM (because rate(incoming) >> rate(processing)).


-- tcr (address@hidden)  ``Ho chresim'eidos uch ho poll'eidos sophos''