Revisit patch for Bug#11935?

Revisit patch for Bug#11935?

[Ravi Desai]

Hello all,

Lines 294-297 of src/lisp.h on master refer to Bug#11935, which 
discusses including changes submitted for that bug-report once Debian 
stable has a GCC >= 4.8

https://packages.debian.org/stable/devel/ says GCC is at 4.9.2-2.

Is this now a good time to revisit the original patch submitted for 
https://debbugs.gnu.org/cgi/bugreport.cgi?bug=11935 ?

Thanks

Re: Revisit patch for Bug#11935?

[Paul Eggert]

[-- Attachment #1: Type: text/plain, Size: 371 bytes --]

On 12/04/2016 04:07 PM, Ravi Desai wrote:
> Is this now a good time to revisit the original patch submitted for 
> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=11935 ? 

I would guess not. As I recall, Eli is still using GCC 3, and I suspect 
some other developers are still using pre-4.8 GCC. Let's kick this can 
down the road a while more. I installed the attached.


[-- Attachment #2: 0001-src-lisp.h-Update-comment-to-kick-Og-down-the-road.patch --]
[-- Type: text/x-patch, Size: 909 bytes --]

From 02055b6de9df82a68d24ee38e7d19e93191938c2 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sun, 4 Dec 2016 18:19:01 -0800
Subject: [PATCH] * src/lisp.h: Update comment to kick -Og down the road.

---
 src/lisp.h | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/src/lisp.h b/src/lisp.h
index 94f1152..7dd9145 100644
--- a/src/lisp.h
+++ b/src/lisp.h
@@ -292,9 +292,8 @@ error !;
    used elsewhere.
 
    FIXME: Remove the lisp_h_OP macros, and define just the inline OP
-   functions, once most developers have access to GCC 4.8 or later and
-   can use "gcc -Og" to debug.  Maybe in the year 2016.  See
-   Bug#11935.
+   functions, once "gcc -Og" (new to GCC 4.8) works well enough for
+   Emacs developers.  Maybe in the year 2020.  See Bug#11935.
 
    Commentary for these macros can be found near their corresponding
    functions, below.  */
-- 
2.7.4

Re: Revisit patch for Bug#11935?

[Colin Baxter]

On Sun, Dec 04 2016, Paul Eggert wrote:

> On 12/04/2016 04:07 PM, Ravi Desai wrote:
>> Is this now a good time to revisit the original patch submitted for
>> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=11935 ? 
>
> I would guess not. As I recall, Eli is still using GCC 3, and I
> suspect some other developers are still using pre-4.8 GCC. Let's kick
> this can down the road a while more. I installed the attached.

For what it's worth, Debian is only on gcc 4.9 for Jessie. Wheezy has
4.7 or 4.6.

Re: Revisit patch for Bug#11935?

[Ravi Desai]

On 2016-12-04 09:21 PM, Paul Eggert wrote:
> On 12/04/2016 04:07 PM, Ravi Desai wrote:
>> Is this now a good time to revisit the original patch submitted for 
>> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=11935 ? 
>
> I would guess not. As I recall, Eli is still using GCC 3, and I 
> suspect some other developers are still using pre-4.8 GCC. Let's kick 
> this can down the road a while more. I installed the attached.
>

Sounds good.  Thanks for the update.

Re: Revisit patch for Bug#11935?

[Eli Zaretskii]

> From: Paul Eggert <eggert@cs.ucla.edu>
> Date: Sun, 4 Dec 2016 18:21:31 -0800
> 
> On 12/04/2016 04:07 PM, Ravi Desai wrote:
> > Is this now a good time to revisit the original patch submitted for 
> > https://debbugs.gnu.org/cgi/bugreport.cgi?bug=11935 ? 
> 
> I would guess not. As I recall, Eli is still using GCC 3

No, those days are long gone.  5.3.0 is what I have here now.

> and I suspect some other developers are still using pre-4.8
> GCC. Let's kick this can down the road a while more. I installed the
> attached.

Thanks.

I think this is related not only to the GCC version, but also to
whether developers use -Og when building development versions of
Emacs.  I'm still using -O0, and only switch to -Og when I build
pretests (and switch to -O2 when building the released tarballs).
That's because I've found that -Og still makes debugging more
difficult compared to -O0.  For example, many values appear as
"optimized out" in the backtraces, which sometimes requires me to go
up the stack very far to find the value.  (-Og is actually -O1 with a
couple of more aggressive optimizations disabled, so what I see is not
surprising.)

So the question is how many people who customary debug on the C level
build with -O0, regardless of their compiler version.  Could people
who read this please tell what they use?

Re: Revisit patch for Bug#11935?

[martin rudalics]

 > So the question is how many people who customary debug on the C level
 > build with -O0, regardless of their compiler version.  Could people
 > who read this please tell what they use?

On GNU/Linux I have an -O0 version for every toolkit which means more
than 20 (GTK, Lucid, Motif, no-toolkit, no-X, Cairo, GNUStep, GTK with
Lucid scroll bars times three for release, master, and master with my
modifications) builds.  Add three -O0 builds on Windows.  None of all
these builds can be used for editing purposes, so I also have three -O3
builds - two for GTK and one on Windows for doing normal work.

martin

Re: Revisit patch for Bug#11935?

[Paul Eggert]

On 12/05/2016 07:47 AM, Eli Zaretskii wrote:
> -Og still makes debugging more
> difficult compared to -O0.  For example, many values appear as
> "optimized out" in the backtraces, which sometimes requires me to go
> up the stack very far to find the value.

If those "optimized out"s are an obstacle we should report them to the 
GCC developers. I debugged Emacs with -Og, found an instance of 
"optimized out", turned it into a small test case, and filed a bug 
report with the GCC developers here:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78685

Please feel free to add to it, or to report more bugs in this area.

Re: Revisit patch for Bug#11935?

[Stefan Monnier]

> If those "optimized out"s are an obstacle we should report them to the GCC
> developers.

Really?  Last I tried -Og they were so pervasive that it seemed very
clearly like a conscious decision on the part of the GCC maintainers.


        Stefan "using -O0 with most/all checks enabled, all the time"

Re: Revisit patch for Bug#11935?

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Date: Mon, 05 Dec 2016 12:44:23 -0500
> 
> > If those "optimized out"s are an obstacle we should report them to the GCC
> > developers.
> 
> Really?  Last I tried -Og they were so pervasive that it seemed very
> clearly like a conscious decision on the part of the GCC maintainers.

Same here.  Let's see how the GCC developers respond to Paul's bug
report.

>         Stefan "using -O0 with most/all checks enabled, all the time"

Right.