Re: [PATCH] Build process uses script to add CFI directives to x86 asm

Re: [PATCH] Build process uses script to add CFI directives to x86 asm

[Alex Dowad]

> configure already uses its own echo definition

> so ash's echo should not be used

> how did the config script end up calling ash's echo?

OK, thanks for pointing this out. I shouldn't have said that the problem was
with ash's echo.

The original configure script worked when I ran it on Linux Mint + bash, and didn't
work on Alpine Linux + BusyBox ash. I didn't notice echo() in the configure script,
did some experimentation with bash's echo and ash's, found a difference which seemed
to explain the observed problem, and worked around it.

Now it looks like the issue was something else. I can't seem to duplicate the problem
right this moment; however, the configure script as it is right now works on both
platforms.

Mysteries, mysteries...

AD

Re: Re: [PATCH] Build process uses script to add CFI directives to x86 asm

[Raphael Cohn]

[-- Attachment #1: Type: text/plain, Size: 1469 bytes --]

Writing portable shell script is a little hard, but I've found a good
approach is to make something ancient like pdksh a symlink for /bin/sh and
even /bin/bash - typically in a small VM. This makes it much more likely
that your code will work without big changes on AIX, BSDs, etc. The other
way is to use dash, not BusyBox ash, as BusyBox ash includes a few bashisms
depending on the configure options at build time.

If you're looking for cross-script ways of solving problems, then feel free
to nab implementations from my shellfire repo (
https://github.com/shellfire-dev/shellfire et al).


On 19 May 2015 at 21:52, Alex Dowad <alexinbeijing@gmail.com> wrote:

> > configure already uses its own echo definition
>
> > so ash's echo should not be used
>
> > how did the config script end up calling ash's echo?
>
> OK, thanks for pointing this out. I shouldn't have said that the problem
> was
> with ash's echo.
>
> The original configure script worked when I ran it on Linux Mint + bash,
> and didn't
> work on Alpine Linux + BusyBox ash. I didn't notice echo() in the
> configure script,
> did some experimentation with bash's echo and ash's, found a difference
> which seemed
> to explain the observed problem, and worked around it.
>
> Now it looks like the issue was something else. I can't seem to duplicate
> the problem
> right this moment; however, the configure script as it is right now works
> on both
> platforms.
>
> Mysteries, mysteries...
>
> AD
>

[-- Attachment #2: Type: text/html, Size: 2057 bytes --]

Re: Re: [PATCH] Build process uses script to add CFI directives to x86 asm

[u-wsnj]

On Tue, May 19, 2015 at 06:22:24PM +0200, Alex Dowad wrote:
> work on Alpine Linux + BusyBox ash. I didn't notice echo() in the configure script,
> did some experimentation with bash's echo and ash's, found a difference which seemed
> to explain the observed problem, and worked around it.
> 
> Now it looks like the issue was something else. I can't seem to duplicate the problem
> right this moment; however, the configure script as it is right now works on both
> platforms.
> 
> Mysteries, mysteries...

I think there was some "echo" i some Makefile
(even though the use which I remember looked safe
it could possibly profit from becoming printf?)

Rune

Re: [PATCH] Build process uses script to add CFI directives to x86 asm

[Alex Dowad]

> Writing portable shell script is a little hard, but I've found a good
> approach is to make something ancient like pdksh a symlink for /bin/sh and
> even /bin/bash - typically in a small VM. This makes it much more likely
> that your code will work without big changes on AIX, BSDs, etc. The other
> way is to use dash, not BusyBox ash, as BusyBox ash includes a few bashisms
> depending on the configure options at build time.

Just tried swapping all references to /bin/sh to /bin/pdksh instead -- the
config script ran just fine and musl was built correctly (with the added CFI
data). Same with /bin/dash.

Thanks for the idea -- I will keep it in mind next time I am writing shell.

Maintainers: is there still anything blocking this patch from getting merged
in? Please let me know so I can fix any remaining problems.

Thanks,
Alex Dowad

Re: Re: [PATCH] Build process uses script to add CFI directives to x86 asm

[Szabolcs Nagy]

* Alex Dowad <alexinbeijing@gmail.com> [2015-05-20 19:08:19 +0200]:
> 
> Maintainers: is there still anything blocking this patch from getting merged
> in? Please let me know so I can fix any remaining problems.
> 

all your mails start a new thread because they don't have
the In-Reply-To: header set up properly (at least that's
how they appear to me) so it's hard to follow the current
state..

i think there are issues
- as wrapping (we did not yet agree about the cleanest way)
- %.sub rules are not yet handled
- i think the \s \w should be fixed (at least mawk gets them wrong)
- clang

otherwise it should be ok
(build system changes are always painful)