Caret escaping (was: Re: make'ing make-3.18beta4 under mingw/msys...

[Alessandro Vesely]

Paul D. Smith wrote:
> %% KJKHyperion <address@hidden> writes:
>
>   k> all of them, with the possible exclusion of the change to sh_chars_dos 
>   k> (IIRC that controls which strings need to be quoted, but I'm not 100% 
>   k> sure. If that's true, the change is correct and needed).
>
> No; the sh_chars_dos string controls whether GNU make runs the command
> script using the "fast path" or the "slow path".

Thus, in case the "slow path" is taken, the LINE command should be replaced with
SHELL FLAG ESCAPED-LINE, where:
 SHELL may be either a unixy_shell, cmd.exe or the old command.com,
 FLAG should be -c for unixy_shell or /C otherwise, and
 ESCAPED-LINE should be escaped using backslashes for unixy_shell,
   carets for cmd.exe, or
   no escape at all for the old command.exe.
Then there is a batch_mode_shell flag, which is usually set for cmd.exe
and causes a NOT-SO-ESCAPED-LINE to be written to a temporary batch file.
Should we revise those specs in order to make a smarter use of cmd.exe?

Althought that patch also allowed -jN to be specified w/o unixy_shell,
after Paul's clarifications it seems the "-j"-change does not depend on 
escaping.

Typically, smarter escaping might be required if the command is, say,

   egrep ^foo $^ | something

In that case, cmd.exe eats the caret unless it is escaped as ^^foo;
command.com does not. As a workaround one can escape the command in
the makefile, which then cannot be ported to windows 98.