Re: [PATCH] Add shell-quasiquote.

[David Kastrup]

Random832 <address@hidden> writes:

> David Kastrup <address@hidden> writes:
>> Eli Zaretskii <address@hidden> writes:
>>> OK, but that still boils down to binding some more variables.  If we
>>> want to help users with these factoids, we could have a small database
>>> of the known Posix shells and their requirements.
>
> I couldn't find an easy way to do let-style binding of environment
> variables, though I didn't look very hard. And the point of what I was
> suggesting is to let *packages* do this, not *users*, so the "database"
> would have to be part of code that decides what to actually do.
>
>> I think that's overdoing it with regard to shell-quote-argument and
>> friends.  We don't need a full POSIX shell, just something with the most
>> basic quoting conventions of it.  /bin/sh should be fine here.
>
> I think there might be a basic miscommunication here. If you don't
> care about being able to execute arbitrary POSIX shell commands as
> well as is possible on a given system, what's even the point of
> *having* functions like shell-quote-argument, shell-command, etc,
> instead of call-process?

Executing _basic_ POSIX shell commands with arbitrary correctly quoted
arguments?  "arbitrary POSIX shell commands" is just too ambitious of a
goal.

Please read
<URL:https://www.gnu.org/software/autoconf/manual/autoconf-2.64/html_node/Portable-Shell.html#Portable-Shell>
or
<URL:info:autoconf#Portable%20Shell>
M-x (info "(autoconf) Portable Shell")
for an overview of the terror awaiting you.

> In these discussions I've been starting from the assumption that these
> functions actually have a point and that being able to execute complex
> shell commands (which may use advanced redirection, if/case/for/etc,
> parameter expansion, and so forth) from within elisp scripts is a
> desirable feature.

This bears absolutely no relation to shell-quote-argument as
shell-quote-argument has only a very small well-described job.

>> That's all the guarantee you get for calling commands/scripts with
>> `system'.  I don't think we should require more than that or try
>> providing some guarantees in that regard.
>
> On an actual POSIX system, it's guaranteed that "sh" from PATH (not
> necessarily /bin/sh) will be a POSIX shell, and that 'system' will
> execute it and not some other shell. (AIUI this is generally still
> implemented by a hardcoded path, just one which is not necessarily
> /bin/sh)

Except that "actual POSIX systems" don't actually exist a whole lot.
It's all approximations.  And it's totally irrelevant regarding the job
shell-quote-argument has to do.

-- 
David Kastrup