[PATCH] D33327: Use linker script commands in writeMapFile

[Peter Smith via llvm-commits]

Thanks for the clarification,

I'll attempt to split up the above reviews as best I can, and refer
back to the original change when wider context is needed. I can
probably help out in moving OutputSections.clear() up, but I'd want to
avoid conflicting or duplicating what you are looking at yourself.
Hopefully I'll be able to split up the parts of the above patches that
move fabcricateDefaultCommands() further up, and will try and work on
what needs to be done to move them further (currently
fabricateDefaultCommands depends on the PHDRS creation).

Peter

On 20 May 2017 at 00:08, Rafael Avila de Espindola
<rafael.espindola at gmail.com> wrote:
> Peter Smith <peter.smith at linaro.org> writes:
>
>> Thanks very much for the patch, looks good to me too.
>>
>> I've recently submitted https://reviews.llvm.org/D33239 and updated
>> https://reviews.llvm.org/D32799 . Together these move
>> fabricateDefaultCommands() earlier and have SHF_LINK_ORDER and Thunks
>> operate on InputSectionDescriptions. If there is anything I can do to
>> progress these, please let me know?
>
> These changes are hard to review since they try to do a lot in one go.
>
> We know that the core issue is that we have two representations for the
> linker state. Lets try to fix that first.
>
> I will be trying to avoid using the non-linker script representation
> once fabricateDefaultCommands is done by moving the
> OutputSections.clear() earlier bit by bit.
>
> The other part is moving what efectivelly is
>
>   if (!Config->Relocatable) {
>      if (!Script->Opt.HasSections) {
>       fixSectionAlignments();
>       Script->fabricateDefaultCommands();
>     }
>     Script->synchronize();
>  }
>
> Earlier a bit at a time until it is early enough that we can drop Script->synchronize().
>
> Cheers,
> Rafael