Re: [PATCH] sim: common: modernize gennltvals.sh

[Simon Marchi via Gdb-patches <gdb-patches@sourceware.org>]

On 2021-01-18 12:52 p.m., Mike Frysinger wrote:
> On 18 Jan 2021 12:19, Simon Marchi via Gdb-patches wrote:
>> On 2021-01-18 12:13 p.m., Mike Frysinger wrote:
>>> i had run shellcheck and the only warnings (about unquoted expansion)
>>> i didn't think we worth fixing because of the limited scope of the
>>> script.  the variable in question isn't accepting user input, it's
>>> operating on fixed inputs, and if we restrict ourselves to POSIX
>>> shell (which i think we do), then our options are limited, and imo
>>> the alternatives make it harder to read/understand.
>>
>> Even considering this, I think it's worth just quoting the variables
>> and getting rid the warnings.  It's trivial, and it would make any
>> future (more important) warning more apparent.
> 
> to be clear, it isn't a style issue, it's a correctness issue.
> adding the quotes will break the script.
> 
> here's the warning:
>   printf '#include <%s>\n' ${files}
>                            ^------^ SC2086: Double quote to prevent globbing and word splitting.
> 
> this is because the code passes in multiple files to process as an arg:
>   gentvals "" errno ... "errno.h sys/errno.h" ...
> 
> then we use it like:
>   files=$4
> ...
>   for f in ${files}; do
> ...
>   printf '#include <%s>\n' ${files}
> 
> unquoted, we get:
> #include <errno.h>
> #include <sys/errno.h>
> 
> quoted, we get:
> #include <errno.h sys/errno.h>

Wait, what.  This printf call gives you multiple lines, and does not
tell you "you have too many arguments for the format string"?  This
is... surprising.

To produce that output, I would have expected a for loop:

for f in ${files}; do
  printf '#include <%s>\n' "$f"
done

> 
> our options are limited with POSIX shell:
> * use arrays ... POSIX shell only has one builtin array: the args.
>   so we'd rework the func API to pass in multiple files, and we'd
>   operate on $@ by shifting it and iterating.

Ok, I'm used to using bash arrays, I forgot they weren't standard.

> * change the shell runtime env by disabling path expansion.  that
>   would mitigate the path expansion (which doesn't happen here as
>   we know the inputs are all alphanumeric/periods), but still would
>   have word splitting because we want that.
> 
> shell is just a bad programming language.  but i think the tree
> specifically constrains itself to it for portability.  i don't know
> what, if any, policies we have about using any other language.
> -mike

I think it would be perfectly fine to require bash or python.

Simon