Re: [PATCH 0/4] Makefile: micro-optimize light non-test builds

From: Jeff King <peff@peff.net>
Date: Jan 27, 2021, 04:34 UTC
Message-ID: <YBDtO82x5sBAs/6L@coredump.intra.peff.net>
In-Reply-To: <xmqq5z3jyxa7.fsf@gitster.c.googlers.com>

On Tue, Jan 26, 2021 at 05:38:08PM -0800, Junio C Hamano wrote:

Show 5 quoted lines

> The steps 2/4 and 3/4 did look like a useful feature, but I wonder
> why we even need to introduce NO_TEST_TOOLS in the first place.
> Wouldn't it be more natural to arrange them to be built by making
> "test::" target depend on them?  IOW, why do we need to have "all::"
> (our default) target depend on them?

Hmm. That is definitely more logical, and giving "make" more information to make a good decision about what is needed. I do wonder if it would be annoying in two cases, though:

  - people trigger the tests in other ways besides "make test". For
    instance, "make && cd t && make" works, as does just
    "make && cd t && ./t1234". With a more clever Makefile, those would
    fail (or worse, run out-of-date versions of the helpers, producing
    confusing results).

  - during refactoring, I often compile-test as I go (i.e., run "make"
    to see which callers still need changed, then fix them, repeat).
    If that didn't catch test helpers, then I'd think I was done and get
    bit later by "make test" trying to build more code. Not the end of
    the world, but a minor annoyance.

So I think even though I'd argue that giving "make" that extra dependency information is "more correct", we are fighting uphill against existing behavior, as well as things that make doesn't know (like that I expect to be ready to run tests as long as "make all" has finished).

-Peff