Re: git fails with a broken pipe when one quits the pager

Am 03.02.21 um 03:48 schrieb Ævar Arnfjörð Bjarmason:

On Mon, Feb 01 2021, Johannes Sixt wrote:

quoted

Am 01.02.21 um 16:44 schrieb Ævar Arnfjörð Bjarmason:

quoted

On Mon, Feb 01 2021, Vincent Lefevre wrote:

quoted

On 2021-02-01 13:10:21 +0100, Ævar Arnfjörð Bjarmason:

quoted

So we've got the SIGPIPE to indicate the output wasn't fully
consumed.

But the user doesn't care: he quit the pager because he didn't
need more output. So there is no need to signal that the output
wasn't fully consumed. The user already knew that before quitting
the pager!

As noted above, this is assuming way too much about the functionality of
the pager command. We can get a SIGPIPE without the user's intent in
this way. Consider e.g. piping to some remote system via netcat.

That assumption is warranted, IMO. Aren't _you_ stretching the meaning
of "pager" too far here? A pager is intended for presentation to the
user. If someone plays games with it, they should know what they get.

FWIW I replied to this in
https://lore.kernel.org/git/87r1lxeuoj.fsf@evledraar.gmail.com/ (local)

Whatever anyone thinks of the virtues of passing down SIGHUP having e.g
a nc to a remote box be your pager isn't all that unusual.

A pager in any form is fair game. That point is that it is an
*interactive* form of presentation. But you should not use git's pager
as data post-processing facility; that would stretch the meaning of
"pager" too far, and we do not have cater for such abuse of the feature.

-- Hannes