Should "git help" respect the 'pager' setting?

I have my global git config pager set to 'cat', but when I do a "git help <command>", it still uses a pager. This is especially irksome in emacs shell buffers, where I am most of the time. I know I can do a M-x man -> git-<whatever>, but wondered if this was a bug or user error. ("git --no-pager help <command>" does the same.)

12:31 [mcampbell] /tmp  % git --no-pager help log
WARNING: terminal is not fully functional
-  (press RETURN)

Michael Campbell <michael.campbell@gmail.com> writes:

"git help foo" just calls "man git-foo" by default, so what happens is the same as if you called "man git-foo" by hand. Git does not have much control over what man will do, it could probably call "man -P $pager" when the Git pager is set, but I'd find it a bit weird.

If you're an Emacs user, you can read about man.viewer and set it to woman, or set PAGER=cat when inside Emacs.

I personally run M-x git-foo RET, and never run "git help".

Matthieu Moy wrote:

It just needs to set $PAGER or $MANPAGER before the exec(), no? I would argue that it should do this. $GIT_PAGER works everywhere else, but obviously man has no knowledge about it.

I just learnt about man.viewer. There's a small problem with it though: why is there no option for Emacs man corresponding to Emacs woman?

M-x man git-foo RET, you mean? My style is slightly different: I love typing out 'man git log' on the terminal (dashless); I get it to open in an Emacs buffer using this hack:

function man_ () {
	emacsclient -e "(man \"$*\")" 2>&1 >/dev/null || man "$*"
}

alias man=man_

Ramkumar Ramachandra <artagnon@gmail.com> writes:

Yes, that should do the same as "man -P".

I find it a bit weird that Git sets the configuration for external commands, but it may make sense. No strong opinion here.

I guess because no one implemented it ;-).

Yes, sorry.

Matthieu Moy wrote:

I don't mean a setenv() kind of thing: how would we unset it after that? Perhaps something like execvpe(), passing in the environment as an argument?

On Thu, May 30, 2013 at 10:38:59PM +0530, Ramkumar Ramachandra wrote:

Overriding PAGER might make sense, but I'd be quite annoyed if Git decided to override MANPAGER without providing some way to override it.

If a user sets MANPAGER then it's because they want a specific pager when reading man pages - invoking man through "git help" shouldn't cause it to behave differently in this case.

John Keeping <john@keeping.me.uk> writes:

Running "man" is the last thing we do before exitting "git help" and exec_man_man() does seem to do execlp(), so I do not see it as an issue.

Hmm, see below.

True, and for that matter, if the user sets PAGER to use with non-Git (including manual pages) and also sets GIT_PAGER to use specifically with Git, the user would be annoyed if we passed formatted manual pages to GIT_PAGER by exporting PAGER (or driving "man" with its "-P" option).