zsh: fix performance issues related to PATH

[samhocevar]

When $PATH contains / and command foo is run, zsh will try to execute //foo instead of /foo, which will attempt to contact a network share on server foo, causing DNS and CIFS requests that will greatly slow down the shell. And if /foo actually exists, it will not be found.

As a workaround, we do not insert the additional '/' if the path element already ends with a slash.

Another workaround for the users would be to put /. instead of / in $PATH.

[samhocevar]

A very easy repro example (using $RANDOM to bypass hashing):

% for d in / /.; do echo PATH=$d; for _ in 1 2 3; do time PATH=$d unknowncommand$RANDOM 2>/dev/null; done; done
PATH=/
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.01s user 0.00s system 1% cpu 1.303 total
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.03s user 0.00s system 2% cpu 1.306 total
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.01s user 0.01s system 2% cpu 1.310 total
PATH=/.
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.00s user 0.00s system 0% cpu 0.010 total
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.00s user 0.00s system 0% cpu 0.010 total
PATH=$d unknowncommand$RANDOM 2> /dev/null  0.00s user 0.00s system 0% cpu 0.010 total
%

[Biswa96]

Should it be reported to upstream zsh project?

[samhocevar]

I don’t really know. Do they care about the Cygwin emulation layer? Because it’s Cygwin that causes //foo to not be the same as /foo like on most Unix systems, so they aren’t wrong in making the assumption.

[lazka]

Just wondering, why do you have "/" in PATH?

I haven't tested it but "///" might also work as a workaround, as that is equal to / according to POSIX, while "//" isn't.