Re: [PATCH 1/7] Redirect grep's stderr top null too.

From: Eric Sunshine <sunshine@sunshineco.com>
Date: May 19, 2020, 16:09 UTC
Message-ID: <CAPig+cTCbBE4Xj7LGJDqz4pcRsR-ZAF-ekcMnZ_cDuzgmEnxcw@mail.gmail.com>
In-Reply-To: <20200518153025.GD1980@danh.dev>

On Mon, May 18, 2020 at 11:30 AM Đoàn Trần Công Danh <congdanhqx@gmail.com> wrote:

Show 9 quoted lines

> On 2020-05-19 00:29:47+1000, Darren Tucker <dtucker@dtucker.net> wrote:
> > $ ./t5703-upload-pack-ref-in-want.sh
> > sed: There are too many commands for the s/\n// function.
>
> I think this was introduced at 878f988350 (t/test-lib: teach
> --chain-lint to detect broken &&-chains in subshells, 2018-07-11)
>
> The chainlint.sed is too complicated for a mortal like me to
> understand, I added Eric to Cc.

That's a rather weird error message; seems like that 'sed' is somewhat broken.

Back when Ævar was trying to get chain-lint to work on some really old and broken platforms, it was ultimately decided (if I recall correctly) that it wasn't worth the effort, and that chain-lint should simply be disabled via GIT_TEST_CHAIN_LINT=0 or --no-chain-lint on those platforms.

After all, chain-lint exists only to ferret out a specific problem in _newly-written_ tests (it's testing the tests), not to ferret out problems in Git functionality (that's what the test suite itself is for). So it's not a great loss to disable chain-lint on an old or broken platform on which it is unlikely someone will be developing _new_ tests. (And, even if someone does write a test on such a platform, &&-chain breakage will be discovered soon enough once the patch is posted to the mailing list and someone runs it on a non-broken platform.)