objcopy: A potentially new option --ignore-symbol(s)

objcopy: A potentially new option --ignore-symbol(s)

[Ilya Kuklin]

Hello,

There is an option in the making for llvm-objcopy called --ignore-symbol(s).
We would like to have some feedback about the name of the option
and its functionality.

The functionality of this option is to ignore the specified symbols when 
executing other options that can change the symbol's name, binding or 
visibility, so that their parameters remain unchanged.

For example, when executing an option --localize-hidden, specifying an 
existing hidden symbol with --ignore-symbol will keep its binding from 
being changed.

Purpose:
This option is needed mostly to support cases where there is a need to 
change parameters of a vast number of symbols with exception of a few 
specific symbols. Writing a negative pattern is much simpler than many 
positive patterns that avoid a certain case. Plus, since there is an option 
--keep-symbol that does exactly this but for removing symbols, there 
should also be one that keeps symbols' parameters from being changed.

If an option with the same functionality is going to be introduced at some 
point to binutils objcopy, please consider naming it the same way or
suggest an alternative.

Feel free to join the discussion:
https://github.com/llvm/llvm-project/pull/80873

--
Thanks,
Ilya Kuklin

Re: objcopy: A potentially new option --ignore-symbol(s)

[Jan Beulich]

On 22.02.2024 18:51, Ilya Kuklin wrote:
> There is an option in the making for llvm-objcopy called --ignore-symbol(s).
> We would like to have some feedback about the name of the option
> and its functionality.
> 
> The functionality of this option is to ignore the specified symbols when 
> executing other options that can change the symbol's name, binding or 
> visibility, so that their parameters remain unchanged.

I think e.g. "skip" would be more appropriate than "ignore".

> For example, when executing an option --localize-hidden, specifying an 
> existing hidden symbol with --ignore-symbol will keep its binding from 
> being changed.

Question is how useful a single negative option is when perhaps you want
to do multiple adjustments at the same time, with different sets needing
exclusion. Imo separate negative options (e.g. --no-localize-hidden)
would be more suitable in the long run.

Jan

> Purpose:
> This option is needed mostly to support cases where there is a need to 
> change parameters of a vast number of symbols with exception of a few 
> specific symbols. Writing a negative pattern is much simpler than many 
> positive patterns that avoid a certain case. Plus, since there is an option 
> --keep-symbol that does exactly this but for removing symbols, there 
> should also be one that keeps symbols' parameters from being changed.
> 
> If an option with the same functionality is going to be introduced at some 
> point to binutils objcopy, please consider naming it the same way or
> suggest an alternative.
> 
> Feel free to join the discussion:
> https://github.com/llvm/llvm-project/pull/80873
> 
> --
> Thanks,
> Ilya Kuklin

Re: objcopy: A potentially new option --ignore-symbol(s)

[Ilya Kuklin]

>> There is an option in the making for llvm-objcopy called --ignore-symbol(s).
>> We would like to have some feedback about the name of the option
>> and its functionality.
>>
>> The functionality of this option is to ignore the specified symbols when
>> executing other options that can change the symbol's name, binding or
>> visibility, so that their parameters remain unchanged.
> 
> I think e.g. "skip" would be more appropriate than "ignore".

Thank you for the feedback! I will rename the options to --skip-symbol(s),
it does sound more fitting indeed.

>> For example, when executing an option --localize-hidden, specifying an
>> existing hidden symbol with --ignore-symbol will keep its binding from
>> being changed.
> 
> Question is how useful a single negative option is when perhaps you want
> to do multiple adjustments at the same time, with different sets needing
> exclusion. Imo separate negative options (e.g. --no-localize-hidden)
> would be more suitable in the long run.
> 
> Jan

It was made as an analogy to the existing --keep-symbol(s) option, which is
a single options that works for all other options that can strip symbols.
The idea was to also have a single option that can work with options that
change symbols' name, binding and visibility in order to keep it simple.
There are a lot of options that do that, so adding a negative option for
every one of them seems like somewhat overcomplicating it. If there is a
need to have different negative patterns for different options, then I feel
like it's a better idea to just run the tool multiple times.

>> Purpose:
>> This option is needed mostly to support cases where there is a need to
>> change parameters of a vast number of symbols with exception of a few
>> specific symbols. Writing a negative pattern is much simpler than many
>> positive patterns that avoid a certain case. Plus, since there is an option
>> --keep-symbol that does exactly this but for removing symbols, there
>> should also be one that keeps symbols' parameters from being changed.
>>
>> If an option with the same functionality is going to be introduced at some
>> point to binutils objcopy, please consider naming it the same way or
>> suggest an alternative.
>>
>> Feel free to join the discussion:
>> https://github.com/llvm/llvm-project/pull/80873