llvm dev - Nov 2017 - Debugging LLVM IR - Reviving the DebugIR pass

Hello all,

I recently felt the need to have debug info at the LLVM IR level (because
my frontend's semantics vastly differ from IR that it doesn't make much
sense to use it). Asm can sometimes be too "low level", so it's
nice to
have the ability to single-step at the IR level.

The pass that used to do this (-debugir) bitrotted and was removed from
tree. I spent the weekend bringing it back: https://reviews.llvm.org/D40207

I'd love to have some help with reviews and tests I should write for this,
because I don't understand the debug info side of LLVM very well.

Also, is there a protocol to have a review be put on the "review
corner"?
I'd be glad if this patch could be listed there :)

Thanks,
~Siddharth.
-- 
Sending this from my phone, please excuse any typos!
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171120/b69202b1/attachment.html>

On Mon, Nov 20, 2017 at 2:32 AM Siddharth Bhat via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hello all,
>
> I recently felt the need to have debug info at the LLVM IR level (because
> my frontend's semantics vastly differ from IR that it doesn't make
much
> sense to use it). Asm can sometimes be too "low level", so
it's nice to
> have the ability to single-step at the IR level.
>
> The pass that used to do this (-debugir) bitrotted and was removed from
> tree. I spent the weekend bringing it back:
> https://reviews.llvm.org/D40207
>
>
I removed the pass because it wasn't maintained and the cost of keeping it
up to date was more than the benefit, so a few questions:

a) Are you planning on maintaining this?
b) Are you going to be doing active work on it?
c) What's the benefit here? What use case?

> I'd love to have some help with reviews and tests I should write for
this,
> because I don't understand the debug info side of LLVM very well.
>
> Also, is there a protocol to have a review be put on the "review
corner"?
> I'd be glad if this patch could be listed there :)
>
>
Are you finding it hard to get a review on the patch?

-eric
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171120/a50aff04/attachment.html>

> I removed the pass because it wasn't maintained and the cost of keeping
it up to date was more than the benefit, so a few questions:
> a) Are you planning on maintaining this?
Yes, I do plan on maintaining this.

> b) Are you going to be doing active work on it?
I do plan on writing additional tests, and maybe adding support for
variables as described in the kaleidoscope docs, yes.

> c) What's the benefit here? What use case?
At least for me, the use case is when you have a miscompile from your
frontend. It's easier to think about things on a LLVM IR level than on the
assembly (at least for me). So, I prefer to have the ability to step
through the LLVM IR in a debugger.
> Are you finding it hard to get a review on the patch?
Yes, I am.

Thanks a lot,
~Siddharth

On Mon, 20 Nov 2017 at 23:57 Eric Christopher <echristo at gmail.com>
wrote:
> On Mon, Nov 20, 2017 at 2:32 AM Siddharth Bhat via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> Hello all,
>>
>> I recently felt the need to have debug info at the LLVM IR level
(because
>> my frontend's semantics vastly differ from IR that it doesn't
make much
>> sense to use it). Asm can sometimes be too "low level", so
it's nice to
>> have the ability to single-step at the IR level.
>>
>> The pass that used to do this (-debugir) bitrotted and was removed from
>> tree. I spent the weekend bringing it back:
>> https://reviews.llvm.org/D40207
>>
>>
> I removed the pass because it wasn't maintained and the cost of keeping
it
> up to date was more than the benefit, so a few questions:
>
> a) Are you planning on maintaining this?
> b) Are you going to be doing active work on it?
> c) What's the benefit here? What use case?
>
>
>> I'd love to have some help with reviews and tests I should write
for
>> this, because I don't understand the debug info side of LLVM very
well.
>>
>> Also, is there a protocol to have a review be put on the "review
corner"?
>> I'd be glad if this patch could be listed there :)
>>
>>
> Are you finding it hard to get a review on the patch?
>
> -eric
>
-- 
Sending this from my phone, please excuse any typos!
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171121/4f8b741b/attachment.html>

On 20 November 2017 at 10:32, Siddharth Bhat via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Hello all,
>
> I recently felt the need to have debug info at the LLVM IR level (because
my
> frontend's semantics vastly differ from IR that it doesn't make
much sense
> to use it). Asm can sometimes be too "low level", so it's
nice to have the
> ability to single-step at the IR level.
>
> The pass that used to do this (-debugir) bitrotted and was removed from
> tree. I spent the weekend bringing it back: https://reviews.llvm.org/D40207
>
> I'd love to have some help with reviews and tests I should write for
this,
> because I don't understand the debug info side of LLVM very well.
>
> Also, is there a protocol to have a review be put on the "review
corner"?
Hi Siddharth, this is documented at http://llvmweekly.org/reviewcorner
which is linked to in each issue. Pasting the guidance here for
convenience:

"""
For inclusion, the following should be true:
* Your patch has gone at least two weeks without substantial review
feedback OR this is your first patch to an LLVM project
* You have updated the patch based on any review comments received so far
* The patch has been compile tested against the current HEAD of the
appropriate LLVM project, and rebased if necessary
* You are willing to write a short two-sentence summary that explains
1) what the patch does, and 2) why people might interested.
"""

Best,

Alex

@Alex: thanks! If I don't get a review this week, I'll put it up :)

@Eric: gentle ping. I replied to the questions you asked. Are there any
other concerns?

Thanks,
Siddharth

On Tue 21 Nov, 2017, 14:40 Alex Bradbury, <asb at asbradbury.org> wrote:
> On 20 November 2017 at 10:32, Siddharth Bhat via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > Hello all,
> >
> > I recently felt the need to have debug info at the LLVM IR level
> (because my
> > frontend's semantics vastly differ from IR that it doesn't
make much
> sense
> > to use it). Asm can sometimes be too "low level", so
it's nice to have
> the
> > ability to single-step at the IR level.
> >
> > The pass that used to do this (-debugir) bitrotted and was removed
from
> > tree. I spent the weekend bringing it back:
> https://reviews.llvm.org/D40207
> >
> > I'd love to have some help with reviews and tests I should write
for
> this,
> > because I don't understand the debug info side of LLVM very well.
> >
> > Also, is there a protocol to have a review be put on the "review
corner"?
>
> Hi Siddharth, this is documented at http://llvmweekly.org/reviewcorner
> which is linked to in each issue. Pasting the guidance here for
> convenience:
>
> """
> For inclusion, the following should be true:
> * Your patch has gone at least two weeks without substantial review
> feedback OR this is your first patch to an LLVM project
> * You have updated the patch based on any review comments received so far
> * The patch has been compile tested against the current HEAD of the
> appropriate LLVM project, and rebased if necessary
> * You are willing to write a short two-sentence summary that explains
> 1) what the patch does, and 2) why people might interested.
> """
>
> Best,
>
> Alex
>
-- 
Sending this from my phone, please excuse any typos!
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171124/78382b43/attachment.html>