[RFC] test: Old shells and physical paths

[RFC] test: Old shells and physical paths

[David Michael]

Hi,

In working on a port, I have to tolerate an ancient shell.  The "cd"
and "pwd" commands don't understand the "-P" flag for physical paths,
as some tests use.  The biggest offender is "cd -P" causing a failure
in t/test-lib.sh (since 1bd9c64), which is sourced by every test
script.

This shell does understand the difference between physical and logical
paths, but the only way to choose which is with a shell option.  The
shell option is named "logical" which is not portable; GNU bash uses
the opposite, "physical".

Does anyone have any suggestions for how to handle this?

Would it be acceptable to instead force the platform's shell option
(if it exists) to always use physical paths for the tests and drop the
"-P" flags?

If not, is it worth maintaining compatibility wrappers for this one
obscure platform?

Thanks.

David

Re: [RFC] test: Old shells and physical paths

[Junio C Hamano]

David Michael <fedora.dm0@gmail.com> writes:

> In working on a port, I have to tolerate an ancient shell.  The "cd"
> and "pwd" commands don't understand the "-P" flag for physical paths,
> as some tests use.  The biggest offender is "cd -P" causing a failure
> in t/test-lib.sh (since 1bd9c64), which is sourced by every test
> script.

Is "here is a nickel, get a better shell" an option?  Running tests
is one thing, but I'd be worried more about scripted Porcelains
broken by a non-POSIX shell if I were you.

> Would it be acceptable to instead force the platform's shell option
> (if it exists) to always use physical paths for the tests and drop the
> "-P" flags?

As a patch to the source files in my tree?  Not likely, even though
I cannot say for sure without looking at how the change would look
like.

Re: [RFC] test: Old shells and physical paths

[David Michael]

Hi,

On Thu, Dec 20, 2012 at 12:17 AM, Junio C Hamano <gitster@pobox.com> wrote:
> Is "here is a nickel, get a better shell" an option?

It is, somewhat.  There is a pre-built port of GNU bash 2.03 for the
platform, but I was trying to see how far things could go with the
OS's supported shell before having to bring in unsupported
dependencies.  Unfortunately, I do not believe the OS fully conforms
to POSIX.1-2001 yet, so that means no "-P" or "-L" without going
rogue.

I'll carry test fixes for this platform locally.

Thanks.

David

Re: [RFC] test: Old shells and physical paths

[David Aguilar]

On Wed, Dec 19, 2012 at 6:28 PM, David Michael <fedora.dm0@gmail.com> wrote:
> Hi,
>
> On Thu, Dec 20, 2012 at 12:17 AM, Junio C Hamano <gitster@pobox.com> wrote:
>> Is "here is a nickel, get a better shell" an option?
>
> It is, somewhat.  There is a pre-built port of GNU bash 2.03 for the
> platform, but I was trying to see how far things could go with the
> OS's supported shell before having to bring in unsupported
> dependencies.  Unfortunately, I do not believe the OS fully conforms
> to POSIX.1-2001 yet, so that means no "-P" or "-L" without going
> rogue.
>
> I'll carry test fixes for this platform locally.

Do you know if the differences are relegated to "cd",
or do other common commands such as awk, grep, sed, mktemp, expr,
etc. have similar issues?

I wonder if it'd be helpful to have a low-numbered test that checks
the basics needed by the scripted Porcelains and test suite.
It would give us an easy way to answer these questions, and could
be a good way to document (in code) the capabilities we expect.
-- 
David

Re: [RFC] test: Old shells and physical paths

[David Michael]

Hi,

On Thu, Dec 20, 2012 at 12:01 AM, David Aguilar <davvid@gmail.com> wrote:
> Do you know if the differences are relegated to "cd",
> or do other common commands such as awk, grep, sed, mktemp, expr,
> etc. have similar issues?

There are almost certainly going to be incompatibilities with other
commands.  The system implemented UNIX95 plus some extensions, then
they began supporting UNIX03/SUSv3/POSIX.1-2001/whatever for certain
commands by using an environment variable to toggle between the
incompatible behaviors.

Their documentation on the UNIX03 commands indicates it is still only
partially supported.  For example: "cp" supports "-L" and "-P", but
"cd" doesn't.

> I wonder if it'd be helpful to have a low-numbered test that checks
> the basics needed by the scripted Porcelains and test suite.
> It would give us an easy way to answer these questions, and could
> be a good way to document (in code) the capabilities we expect.

I'd be in favor of something like this as well.

Thanks.

David


P.S.
In the meantime, I am handling the "cd" situation by replacing "-P"
with "$PHYS" and prepending the following to t/test-lib.sh.
set +o logical >/dev/null 2>&1 || PHYS=-P