bug#19350: #19350 24.4; Incorrect quoting of %-signs for Windows command shell

[Noam Postavsky <npostavs@users.sourceforge.net>]

tag 19350 confirmed
severity 19350 minor
quit

> Looking at the code in subr.el, it is clear that Emacs mishandles
> %-signs when escaping for the Windows command interpreter.
[...]
>
> I deleted the lossage from the report, as there is no point in including
> it, this being a logic error in subr.el. In fact, the buggy code is dead
> on my machine (I run Linux) -- I happened to know about the problem on
> Windows.

Well, it would have been nice to mention which function you're talking
about, and give an example problematic string.  I guess it's
shell-quote-argument:

Evaluating

(let ((process-environment (cl-list* "ca^=with-caret"
                                     "ca=without-caret"
                                     process-environment)))
  (insert (shell-command-to-string
           (format "echo %s %s %s"
                   "%ca%"
                   (shell-quote-argument "%ca%")
                   "%%CD:~,0%ca%%CD:~,0%"))))
gives

without-caret "with-caret" %ca%

On the other hand, this is such an obscure corner case, I'm don't know
if it's even worth fixing.

> A method of handling % that I believe to be fully
> robust is to follow (not precede) each series of % characters with
> %CD:~,0%. This is an expansion that always expands to the empty
> string. It will work provided that:
>
> * Command Extensions are enabled.

I'm not sure if Emacs can assume this.

>
> * It is not possible for cmd.exe to interpret an environment variable
>   whose name is the empty string.

I found that adding "=emptyvar" to process-environment caused
"Spawning child process: exec format error", so I guess this
assumption is safe.