llvm dev - Sep 2013 - [LLVMdev] Ambiguity in LLVM IR for global variable sections

What do we expect from the following IR?

@i2 = constant i32 30, section "INITDATA", align 4
@i1 = global i32 0, section "INITDATA", align 4

Currently, we stick them in one section that isn't writable which is
problematic because @i1 is writable.  If you flip @i1 and @i2 you get one
section which is writable.

It seems like we should do one of the following:
1. Two sections, one writable + one that isn't
2. One section that is writable
3. Give an error

Personally, I like #1 or #2.

-- 
David Majnemer
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130915/29eb8d45/attachment.html>

Seems like #2 won't work for this case:

@l = thread_local global i32 0, section "bar", align 4
@k = global i32 0, section "bar", align 4


On Sun, Sep 15, 2013 at 6:40 PM, David Majnemer <david.majnemer at
gmail.com>wrote:
> What do we expect from the following IR?
>
> @i2 = constant i32 30, section "INITDATA", align 4
> @i1 = global i32 0, section "INITDATA", align 4
>
> Currently, we stick them in one section that isn't writable which is
> problematic because @i1 is writable.  If you flip @i1 and @i2 you get one
> section which is writable.
>
> It seems like we should do one of the following:
> 1. Two sections, one writable + one that isn't
> 2. One section that is writable
> 3. Give an error
>
> Personally, I like #1 or #2.
>
> --
> David Majnemer
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130916/cf78c3bd/attachment.html>

On 15 September 2013 21:40, David Majnemer <david.majnemer at gmail.com>
wrote:
> What do we expect from the following IR?
>
> @i2 = constant i32 30, section "INITDATA", align 4
> @i1 = global i32 0, section "INITDATA", align 4
>
> Currently, we stick them in one section that isn't writable which is
> problematic because @i1 is writable.  If you flip @i1 and @i2 you get one
> section which is writable.
>
> It seems like we should do one of the following:
> 1. Two sections, one writable + one that isn't
> 2. One section that is writable
> 3. Give an error
>
> Personally, I like #1 or #2.
Why not #3? It the frontend wants a constant section it can print @i1
differently. It if wants a writable section it can print @i2
differently.

Cheers,
Rafael

I've come around and believe #3 is the right answer. Better yet, we should
give a way to specify the section attributes somewhere explicitly and only
infer them as a fallback.


On Mon, Oct 7, 2013 at 7:45 AM, Rafael Espíndola <rafael.espindola at
gmail.com
> wrote:
> On 15 September 2013 21:40, David Majnemer <david.majnemer at
gmail.com>
> wrote:
> > What do we expect from the following IR?
> >
> > @i2 = constant i32 30, section "INITDATA", align 4
> > @i1 = global i32 0, section "INITDATA", align 4
> >
> > Currently, we stick them in one section that isn't writable which
is
> > problematic because @i1 is writable.  If you flip @i1 and @i2 you get
one
> > section which is writable.
> >
> > It seems like we should do one of the following:
> > 1. Two sections, one writable + one that isn't
> > 2. One section that is writable
> > 3. Give an error
> >
> > Personally, I like #1 or #2.
>
> Why not #3? It the frontend wants a constant section it can print @i1
> differently. It if wants a writable section it can print @i2
> differently.
>
> Cheers,
> Rafael
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20131010/333fa728/attachment.html>