thr3ads.net - llvm dev - [LLVMdev] [RFC] First-class debug info IR: MDLocation (redux) [Jan 2015]

If this information is useful, please help other people find it:
Share via:

Duncan P. N. Exon Smith

2015-Jan-13 04:30 UTC

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

After splitting `Metadata` apart from `Value`, I'm finally back to the
first step [1] of representing debug info more directly in IR [2]:
`MDLocation`.

[1]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/078173.html
        "First-class debug info IR: MDLocation"
[2]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/077715.html
        "Less memory and greater maintainability for debug info IR"

I've attached two patches which are both essentially commit-ready (just
missing LangRef update, AFAIK), although they just add the class and
assembly/bitcode support.  Once it's in tree, I'll move it into place
underneath `DebugLoc` and `DILocation` and update the world of
testcases, publish a script, and so forth.

Assembly syntax
==============
I've changed the assembly syntax a little since my previous proposal
[1], but I kept the fundamentals the same:

  - Fields use key/value pairs and can be specified in any order.
  - No types (unless you're referencing a `Value`).
  - Some fields are required (in this case, 'scope:').
  - Unspecified fields get sane default values.

Old syntax:

    !6 = metadata MDLocation(line: 43, column: 7, scope: !4)

Now it's:

    !6 = !MDLocation(line: 43, column: 7, scope: !4)

Here's why:

  - Dropped 'metadata' since `Metadata` is now typeless \O/.

  - Added a '!' since it makes the parser simpler (no changes to the
    lexer, since we can unambiguously reuse the `MetadataVar` token),
    and follows the rule that "All metadata are identified in syntax by
    a[n] exclamation point" [3].  (If people are opposed to the extra
    '!' I can update this sentence instead, but it seems like a handy
    mnemonic to maintain.)

[3]: http://llvm.org/docs/LangRef.html#metadata

Where this is heading
====================
See my previous post [1] ;).

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-IR-Add-MDLocation-class.patch
Type: application/octet-stream
Size: 14059 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150112/bceccecd/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-AsmParser-Bitcode-Add-support-for-MDLocation.patch
Type: application/octet-stream
Size: 19200 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150112/bceccecd/attachment-0001.obj>

Robinson, Paul

2015-Jan-13 05:50 UTC

head link

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

Would you mind terribly waiting a week, until after 3.6 is forked?
We're barely caught up with the last big metadata syntax change.
Thanks,
--paulr

From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On
Behalf Of Duncan P. N. Exon Smith
Sent: Monday, January 12, 2015 8:30 PM
To: LLVM Developers Mailing List
Subject: [LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

After splitting `Metadata` apart from `Value`, I'm finally back to the
first step [1] of representing debug info more directly in IR [2]:
`MDLocation`.

[1]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/078173.html
        "First-class debug info IR: MDLocation"
[2]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/077715.html
        "Less memory and greater maintainability for debug info IR"

I've attached two patches which are both essentially commit-ready (just
missing LangRef update, AFAIK), although they just add the class and
assembly/bitcode support.  Once it's in tree, I'll move it into place
underneath `DebugLoc` and `DILocation` and update the world of
testcases, publish a script, and so forth.

Assembly syntax
==============
I've changed the assembly syntax a little since my previous proposal
[1], but I kept the fundamentals the same:

  - Fields use key/value pairs and can be specified in any order.
  - No types (unless you're referencing a `Value`).
  - Some fields are required (in this case, 'scope:').
  - Unspecified fields get sane default values.

Old syntax:

    !6 = metadata MDLocation(line: 43, column: 7, scope: !4)

Now it's:

    !6 = !MDLocation(line: 43, column: 7, scope: !4)

Here's why:

  - Dropped 'metadata' since `Metadata` is now typeless \O/.

  - Added a '!' since it makes the parser simpler (no changes to the
    lexer, since we can unambiguously reuse the `MetadataVar` token),
    and follows the rule that "All metadata are identified in syntax by
    a[n] exclamation point" [3].  (If people are opposed to the extra
    '!' I can update this sentence instead, but it seems like a handy
    mnemonic to maintain.)

[3]: http://llvm.org/docs/LangRef.html#metadata

Where this is heading
====================
See my previous post [1] ;).
_______________________________________________
LLVM Developers mailing list
LLVMdev at cs.uiuc.edu<mailto:LLVMdev at cs.uiuc.edu>        
http://llvm.cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150113/6185285e/attachment.html>

Eric Christopher

2015-Jan-13 18:13 UTC

head link

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

This LGTM. I like the new syntax. Feel free to commit as soon as you have
the LangRef changes handy.

-eric

On Mon Jan 12 2015 at 8:34:18 PM Duncan P. N. Exon Smith <
dexonsmith at apple.com> wrote:
> After splitting `Metadata` apart from `Value`, I'm finally back to the
> first step [1] of representing debug info more directly in IR [2]:
> `MDLocation`.
>
> [1]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/078173.html
>         "First-class debug info IR: MDLocation"
> [2]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/077715.html
>         "Less memory and greater maintainability for debug info
IR"
>
> I've attached two patches which are both essentially commit-ready (just
> missing LangRef update, AFAIK), although they just add the class and
> assembly/bitcode support.  Once it's in tree, I'll move it into
place
> underneath `DebugLoc` and `DILocation` and update the world of
> testcases, publish a script, and so forth.
>
> Assembly syntax
> ==============>
> I've changed the assembly syntax a little since my previous proposal
> [1], but I kept the fundamentals the same:
>
>   - Fields use key/value pairs and can be specified in any order.
>   - No types (unless you're referencing a `Value`).
>   - Some fields are required (in this case, 'scope:').
>   - Unspecified fields get sane default values.
>
> Old syntax:
>
>     !6 = metadata MDLocation(line: 43, column: 7, scope: !4)
>
> Now it's:
>
>     !6 = !MDLocation(line: 43, column: 7, scope: !4)
>
> Here's why:
>
>   - Dropped 'metadata' since `Metadata` is now typeless \O/.
>
>   - Added a '!' since it makes the parser simpler (no changes to
the
>     lexer, since we can unambiguously reuse the `MetadataVar` token),
>     and follows the rule that "All metadata are identified in syntax
by
>     a[n] exclamation point" [3].  (If people are opposed to the extra
>     '!' I can update this sentence instead, but it seems like a
handy
>     mnemonic to maintain.)
>
> [3]: http://llvm.org/docs/LangRef.html#metadata
>
> Where this is heading
> ====================>
> See my previous post [1] ;).
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150113/7252b4c4/attachment.html>

Eric Christopher

2015-Jan-13 18:14 UTC

head link

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

Hi Paul,

I'm not sure I see any reason to wait until after 3.6. More changes are
incoming and it's better to get them over with sooner rather than spread
out the pain.

Thanks!

-eric

On Mon Jan 12 2015 at 9:52:48 PM Robinson, Paul <
Paul_Robinson at playstation.sony.com> wrote:
>  Would you mind terribly waiting a week, until after 3.6 is forked?
>
> We're barely caught up with the last big metadata syntax change.
>
> Thanks,
>
> --paulr
>
>
>
> *From:* llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at
cs.uiuc.edu] *On
> Behalf Of *Duncan P. N. Exon Smith
> *Sent:* Monday, January 12, 2015 8:30 PM
> *To:* LLVM Developers Mailing List
> *Subject:* [LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)
>
>
>
> After splitting `Metadata` apart from `Value`, I'm finally back to the
> first step [1] of representing debug info more directly in IR [2]:
> `MDLocation`.
>
> [1]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/078173.html
>         "First-class debug info IR: MDLocation"
> [2]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-October/077715.html
>         "Less memory and greater maintainability for debug info
IR"
>
> I've attached two patches which are both essentially commit-ready (just
> missing LangRef update, AFAIK), although they just add the class and
> assembly/bitcode support.  Once it's in tree, I'll move it into
place
> underneath `DebugLoc` and `DILocation` and update the world of
> testcases, publish a script, and so forth.
>
> Assembly syntax
> ==============>
> I've changed the assembly syntax a little since my previous proposal
> [1], but I kept the fundamentals the same:
>
>   - Fields use key/value pairs and can be specified in any order.
>   - No types (unless you're referencing a `Value`).
>   - Some fields are required (in this case, 'scope:').
>   - Unspecified fields get sane default values.
>
> Old syntax:
>
>     !6 = metadata MDLocation(line: 43, column: 7, scope: !4)
>
> Now it's:
>
>     !6 = !MDLocation(line: 43, column: 7, scope: !4)
>
> Here's why:
>
>   - Dropped 'metadata' since `Metadata` is now typeless \O/.
>
>   - Added a '!' since it makes the parser simpler (no changes to
the
>     lexer, since we can unambiguously reuse the `MetadataVar` token),
>     and follows the rule that "All metadata are identified in syntax
by
>     a[n] exclamation point" [3].  (If people are opposed to the extra
>     '!' I can update this sentence instead, but it seems like a
handy
>     mnemonic to maintain.)
>
> [3]: http://llvm.org/docs/LangRef.html#metadata
>
> Where this is heading
> ====================>
> See my previous post [1] ;).
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>   _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150113/4526e2cd/attachment.html>

Possibly Parallel Threads

Search for more apparently analagous threads

llvm dev - Jan 2015 - [LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

[LLVMdev] [RFC] First-class debug info IR: MDLocation (redux)

Possibly Parallel Threads