[PATCH] D22693: More strongly separate the PDB reading interfaces and PDB writing interfaces

[Zachary Turner via llvm-commits]

I think part of what we're getting out of this is understanding. If we wait
until we need it for the linker specifically, then we don't have any
experience so we're starting from scratch, and the past few weeks will be
an extra few weeks since we have to do all this anyway.

We also need it for not just the linker. Lldb will need it eventually, and
the dumper needs it no matter what for testing.

Even if we have to make some changes to accommodate LLD's needs, at least
we will be starting from something and have a deep knowledge of the format,
ultimately saving us some time.

You mention the code size of LLD being small. Is the size of the linker
important? I thought we are mostly interested in optimizing the speed
On Mon, Jul 25, 2016 at 3:05 PM Rui Ueyama <ruiu at google.com> wrote:

> ruiu added a comment.
>
> At a high level, it looks good to me, but as I probably mentioned earlier,
> I have a little concern that we are having an API for PDB that is too
> complete at this moment. The COFF linker is 7000 lines long in total, and
> if we keep this pace, we'll probably end up having a PDB library which is
> larger than the linker. Even if we'd eventually need it, I'm more
> comfortable if we do as-needed basis.
>
>
> https://reviews.llvm.org/D22693
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160725/f9f8ddf9/attachment.html>