PATCH: Compress debug sections for Linux/x86 by default

PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

I checked in this patch to compress debug sections for Linux/x86 by
default.

H.J.
---
diff --git a/gas/ChangeLog b/gas/ChangeLog
index 8ab2319..27da2cc 100644
--- a/gas/ChangeLog
+++ b/gas/ChangeLog
@@ -1,3 +1,8 @@
+2014-12-14  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* config/tc-i386.c (flag_compress_debug): Default to compress
+	debug sections for Linux.
+
 2014-12-12  Alan Modra  <amodra@gmail.com>
 
 	* config/tc-ppc.h (md_reg_eh_frame_to_debug_frame): Match current
diff --git a/gas/config/tc-i386.c b/gas/config/tc-i386.c
index b6ac902..0a05c2c 100644
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -33,6 +33,11 @@
 #include "elf/x86-64.h"
 #include "opcodes/i386-init.h"
 
+#ifdef TE_LINUX
+/* Default to compress debug sections for Linux.  */
+int flag_compress_debug = 1;
+#endif
+
 #ifndef REGISTER_WARNINGS
 #define REGISTER_WARNINGS 1
 #endif

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> I checked in this patch to compress debug sections for Linux/x86 by
> default.

Has this been discussed before, or is this something obviously
part of a general direction that was agreed on?

Are things still going to work if GDB is configured without zlib,
or instance?

This is also the type of change that I think is worthy of a NEWS
entry, IMO.

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Sun, Dec 14, 2014 at 6:56 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> I checked in this patch to compress debug sections for Linux/x86 by
>> default.
>
> Has this been discussed before, or is this something obviously
> part of a general direction that was agreed on?

Not on binutils mailing list.  I have been using it for several years
and mentioned it in

https://sourceware.org/ml/libc-alpha/2011-06/msg00041.html

---
Starting from the 2.21.51.0.2 release, binutils fully supports compressed
debug sections.  However, compressed debug section isn't turned on by
default in assembler. I am planning to turn it on for x86 assembler in
the future release, which may lead to the Linux kernel bug messages like

WARNING: lib/ts_kmp.o (.zdebug_aranges): unexpected non-allocatable section.

But the resulting kernel works fine.
----

> Are things still going to work if GDB is configured without zlib,
> or instance?

No.  Zlib must be enabled.  Since I only enabled it in Linux/x86
assembler, I think it is acceptable.

> This is also the type of change that I think is worthy of a NEWS
> entry, IMO.
>

Good idea.  Done.

Thanks.


-- 
H.J.
---
diff --git a/gas/ChangeLog b/gas/ChangeLog
index 27da2cc..b53fcff 100644
--- a/gas/ChangeLog
+++ b/gas/ChangeLog
@@ -2,6 +2,7 @@

  * config/tc-i386.c (flag_compress_debug): Default to compress
  debug sections for Linux.
+ * NEWS: Mention it.

 2014-12-12  Alan Modra  <amodra@gmail.com>

diff --git a/gas/NEWS b/gas/NEWS
index 8a757c5..16035c6 100644
--- a/gas/NEWS
+++ b/gas/NEWS
@@ -1,5 +1,9 @@
 -*- text -*-

+Changes in 2.26:
+
+* --compress-debug-sections is turned on for Linux/x86 by default.
+
 Changes in 2.25:

 * Add support for the AVR Tiny microcontrollers.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> > Are things still going to work if GDB is configured without zlib,
> > or instance?
> 
> No.  Zlib must be enabled.  Since I only enabled it in Linux/x86
> assembler, I think it is acceptable.

FWIW, for an incompatible change like that, it would have been nice if
people had a chance to comment on it before you went ahead and pushed
it as if it was obvious. Your old message mentioning this change is
better than nothing for sure, but even if others agreed then, some
of them might still not be ready for this change.

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Sun, Dec 14, 2014 at 7:49 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> > Are things still going to work if GDB is configured without zlib,
>> > or instance?
>>
>> No.  Zlib must be enabled.  Since I only enabled it in Linux/x86
>> assembler, I think it is acceptable.
>
> FWIW, for an incompatible change like that, it would have been nice if
> people had a chance to comment on it before you went ahead and pushed
> it as if it was obvious. Your old message mentioning this change is
> better than nothing for sure, but even if others agreed then, some
> of them might still not be ready for this change.
>

Let's discuss it now.  We have plenty of time to make the final decision
for binutils 2.26.

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> Let's discuss it now.  We have plenty of time to make the final decision
> for binutils 2.26.

I'll leave the decision to the maintainers, as I don't really have
much data on the usefulness of this feature, vs the (temporary)
disruption it might cause.

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Mon, Dec 15, 2014 at 7:22 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> Let's discuss it now.  We have plenty of time to make the final decision
>> for binutils 2.26.
>
> I'll leave the decision to the maintainers, as I don't really have
> much data on the usefulness of this feature, vs the (temporary)
> disruption it might cause.

We have a year to find it out :-).


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joseph Myers]

On Sun, 14 Dec 2014, H.J. Lu wrote:

> I checked in this patch to compress debug sections for Linux/x86 by
> default.

I don't think it makes sense for something like this to be 
architecture-dependent.  If desired it should be done for GNU/Linux on all 
architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems 
with GNU userspace), if not for all ELF targets.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> On Sun, 14 Dec 2014, H.J. Lu wrote:
>
>> I checked in this patch to compress debug sections for Linux/x86 by
>> default.
>
> I don't think it makes sense for something like this to be
> architecture-dependent.  If desired it should be done for GNU/Linux on all
> architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
> with GNU userspace), if not for all ELF targets.
>

Sure, someone can contribute a patch.


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Thomas Schwinge]

[-- Attachment #1: Type: text/plain, Size: 947 bytes --]

Hi!

On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> > On Sun, 14 Dec 2014, H.J. Lu wrote:
> >
> >> I checked in this patch to compress debug sections for Linux/x86 by
> >> default.
> >
> > I don't think it makes sense for something like this to be
> > architecture-dependent.  If desired it should be done for GNU/Linux on all
> > architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
> > with GNU userspace), if not for all ELF targets.
> 
> Sure, someone can contribute a patch.

I don't like this attitude very much.  We should strive to be a bit more
cooperative within our own community, at least.  As Joseph says, there is
no reason to restrict this change -- which, as Joel has pointed out, has
not been discussed/reviewed/approved before -- to x86 GNU/Linux only.


Grüße,
 Thomas

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 472 bytes --]

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
> Hi!
>
> On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>> > On Sun, 14 Dec 2014, H.J. Lu wrote:
>> >
>> >> I checked in this patch to compress debug sections for Linux/x86 by
>> >> default.
>> >
>> > I don't think it makes sense for something like this to be
>> > architecture-dependent.  If desired it should be done for GNU/Linux on all
>> > architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
>> > with GNU userspace), if not for all ELF targets.
>>
>> Sure, someone can contribute a patch.
>
> I don't like this attitude very much.  We should strive to be a bit more
> cooperative within our own community, at least.  As Joseph says, there is
> no reason to restrict this change -- which, as Joel has pointed out, has
> not been discussed/reviewed/approved before -- to x86 GNU/Linux only.

It is next to impossible to get agreements for all targets.
It is better to start from somewhere.  Each target maintainer
should make his/her own decision.


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joseph Myers]

On Thu, 8 Jan 2015, H.J. Lu wrote:

> On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
> > Hi!
> >
> > On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
> >> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> >> > On Sun, 14 Dec 2014, H.J. Lu wrote:
> >> >
> >> >> I checked in this patch to compress debug sections for Linux/x86 by
> >> >> default.
> >> >
> >> > I don't think it makes sense for something like this to be
> >> > architecture-dependent.  If desired it should be done for GNU/Linux on all
> >> > architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
> >> > with GNU userspace), if not for all ELF targets.
> >>
> >> Sure, someone can contribute a patch.
> >
> > I don't like this attitude very much.  We should strive to be a bit more
> > cooperative within our own community, at least.  As Joseph says, there is
> > no reason to restrict this change -- which, as Joel has pointed out, has
> > not been discussed/reviewed/approved before -- to x86 GNU/Linux only.
> 
> It is next to impossible to get agreements for all targets.
> It is better to start from somewhere.  Each target maintainer
> should make his/her own decision.

My point is that agreement for all targets is not what we want anyway.  
We should make a decision for the GNU system (or for ELF targets in 
general) and then apply that to all targets that are part of the GNU 
system (i.e. all with GNU userspace) (or for ELF targets in general).  If 
an OS maintainer thinks it's bad for their OS, don't enable it there, but 
enabling / disabling on a per-architecture (as opposed to per-OS) basis 
doesn't make sense without a compelling justification.  And we don't have 
many OS maintainers.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: PATCH: Compress debug sections for Linux/x86 by default

[Paul_Koning]

> On Jan 8, 2015, at 1:35 PM, Joseph Myers <joseph@codesourcery.com> wrote:
> 
> On Thu, 8 Jan 2015, H.J. Lu wrote:
> 
>> On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
>>> Hi!
>>> 
>>> On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>>>> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>>>>> On Sun, 14 Dec 2014, H.J. Lu wrote:
>>>>> 
>>>>>> I checked in this patch to compress debug sections for Linux/x86 by
>>>>>> default.
>>>>> 
>>>>> I don't think it makes sense for something like this to be
>>>>> architecture-dependent.  If desired it should be done for GNU/Linux on all
>>>>> architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
>>>>> with GNU userspace), if not for all ELF targets.
>>>> 
>>>> Sure, someone can contribute a patch.
>>> 
>>> I don't like this attitude very much.  We should strive to be a bit more
>>> cooperative within our own community, at least.  As Joseph says, there is
>>> no reason to restrict this change -- which, as Joel has pointed out, has
>>> not been discussed/reviewed/approved before -- to x86 GNU/Linux only.
>> 
>> It is next to impossible to get agreements for all targets.
>> It is better to start from somewhere.  Each target maintainer
>> should make his/her own decision.
> 
> My point is that agreement for all targets is not what we want anyway.  
> We should make a decision for the GNU system (or for ELF targets in 
> general) and then apply that to all targets that are part of the GNU 
> system (i.e. all with GNU userspace) (or for ELF targets in general).  If 
> an OS maintainer thinks it's bad for their OS, don't enable it there, but 
> enabling / disabling on a per-architecture (as opposed to per-OS) basis 
> doesn't make sense without a compelling justification.  And we don't have 
> many OS maintainers.

I agree.

At the very most, if you are really concerned about lack of agreement, having the feature in Linux generally may be a valid first step.  But only Linux and then just on ONE target platform doesn’t make any sense.  There is nothing here that is target specific as far as I can tell, so a limitation to x86 is something I do not understand at all.

	paul

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Thu, Jan 8, 2015 at 10:35 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> On Thu, 8 Jan 2015, H.J. Lu wrote:
>
>> On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
>> > Hi!
>> >
>> > On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>> >> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>> >> > On Sun, 14 Dec 2014, H.J. Lu wrote:
>> >> >
>> >> >> I checked in this patch to compress debug sections for Linux/x86 by
>> >> >> default.
>> >> >
>> >> > I don't think it makes sense for something like this to be
>> >> > architecture-dependent.  If desired it should be done for GNU/Linux on all
>> >> > architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
>> >> > with GNU userspace), if not for all ELF targets.
>> >>
>> >> Sure, someone can contribute a patch.
>> >
>> > I don't like this attitude very much.  We should strive to be a bit more
>> > cooperative within our own community, at least.  As Joseph says, there is
>> > no reason to restrict this change -- which, as Joel has pointed out, has
>> > not been discussed/reviewed/approved before -- to x86 GNU/Linux only.
>>
>> It is next to impossible to get agreements for all targets.
>> It is better to start from somewhere.  Each target maintainer
>> should make his/her own decision.
>
> My point is that agreement for all targets is not what we want anyway.
> We should make a decision for the GNU system (or for ELF targets in
> general) and then apply that to all targets that are part of the GNU
> system (i.e. all with GNU userspace) (or for ELF targets in general).  If
> an OS maintainer thinks it's bad for their OS, don't enable it there, but
> enabling / disabling on a per-architecture (as opposed to per-OS) basis
> doesn't make sense without a compelling justification.  And we don't have
> many OS maintainers.
>

We don't even have a Linux OS maintainer for binutils.  What
takes to enable compression by default on Linux?


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Paul_Koning]

> On Jan 8, 2015, at 2:14 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> 
> On Thu, Jan 8, 2015 at 10:35 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>> On Thu, 8 Jan 2015, H.J. Lu wrote:
>> 
>>> On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
>>>> Hi!
>>>> 
>>>> On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>>>>> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>>>>>> On Sun, 14 Dec 2014, H.J. Lu wrote:
>>>>>> 
>>>>>>> I checked in this patch to compress debug sections for Linux/x86 by
>>>>>>> default.
>>>>>> 
>>>>>> I don't think it makes sense for something like this to be
>>>>>> architecture-dependent.  If desired it should be done for GNU/Linux on all
>>>>>> architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
>>>>>> with GNU userspace), if not for all ELF targets.
>>>>> 
>>>>> Sure, someone can contribute a patch.
>>>> 
>>>> I don't like this attitude very much.  We should strive to be a bit more
>>>> cooperative within our own community, at least.  As Joseph says, there is
>>>> no reason to restrict this change -- which, as Joel has pointed out, has
>>>> not been discussed/reviewed/approved before -- to x86 GNU/Linux only.
>>> 
>>> It is next to impossible to get agreements for all targets.
>>> It is better to start from somewhere.  Each target maintainer
>>> should make his/her own decision.
>> 
>> My point is that agreement for all targets is not what we want anyway.
>> We should make a decision for the GNU system (or for ELF targets in
>> general) and then apply that to all targets that are part of the GNU
>> system (i.e. all with GNU userspace) (or for ELF targets in general).  If
>> an OS maintainer thinks it's bad for their OS, don't enable it there, but
>> enabling / disabling on a per-architecture (as opposed to per-OS) basis
>> doesn't make sense without a compelling justification.  And we don't have
>> many OS maintainers.
>> 
> 
> We don't even have a Linux OS maintainer for binutils.  What
> takes to enable compression by default on Linux?

What does it take to enable it generally, independent of OS?  Or at least generally for ELF?  That would seem to be a matter of new machinery in common ELF handling code that’s part of core binutils (BFD, right?).

Also, your comment seems to imply that this feature, which is conceptually completely generic, can only ever be delivered to one target platform on one OS because there aren’t maintainers for other flavors.  That doesn’t seem right.

	paul

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Thu, Jan 8, 2015 at 11:25 AM,  <Paul_Koning@dell.com> wrote:
>
>> On Jan 8, 2015, at 2:14 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>>
>> On Thu, Jan 8, 2015 at 10:35 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>>> On Thu, 8 Jan 2015, H.J. Lu wrote:
>>>
>>>> On Thu, Jan 8, 2015 at 8:18 AM, Thomas Schwinge <thomas@codesourcery.com> wrote:
>>>>> Hi!
>>>>>
>>>>> On Mon, 15 Dec 2014 11:04:01 -0800, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>>>>>> On Mon, Dec 15, 2014 at 10:47 AM, Joseph Myers <joseph@codesourcery.com> wrote:
>>>>>>> On Sun, 14 Dec 2014, H.J. Lu wrote:
>>>>>>>
>>>>>>>> I checked in this patch to compress debug sections for Linux/x86 by
>>>>>>>> default.
>>>>>>>
>>>>>>> I don't think it makes sense for something like this to be
>>>>>>> architecture-dependent.  If desired it should be done for GNU/Linux on all
>>>>>>> architectures (and probably for GNU/Hurd, GNU/kFreeBSD etc. - all systems
>>>>>>> with GNU userspace), if not for all ELF targets.
>>>>>>
>>>>>> Sure, someone can contribute a patch.
>>>>>
>>>>> I don't like this attitude very much.  We should strive to be a bit more
>>>>> cooperative within our own community, at least.  As Joseph says, there is
>>>>> no reason to restrict this change -- which, as Joel has pointed out, has
>>>>> not been discussed/reviewed/approved before -- to x86 GNU/Linux only.
>>>>
>>>> It is next to impossible to get agreements for all targets.
>>>> It is better to start from somewhere.  Each target maintainer
>>>> should make his/her own decision.
>>>
>>> My point is that agreement for all targets is not what we want anyway.
>>> We should make a decision for the GNU system (or for ELF targets in
>>> general) and then apply that to all targets that are part of the GNU
>>> system (i.e. all with GNU userspace) (or for ELF targets in general).  If
>>> an OS maintainer thinks it's bad for their OS, don't enable it there, but
>>> enabling / disabling on a per-architecture (as opposed to per-OS) basis
>>> doesn't make sense without a compelling justification.  And we don't have
>>> many OS maintainers.
>>>
>>
>> We don't even have a Linux OS maintainer for binutils.  What
>> takes to enable compression by default on Linux?
>
> What does it take to enable it generally, independent of OS?  Or at least generally for ELF?  That would seem to be a matter of new machinery in common ELF handling code that’s part of core binutils (BFD, right?).

This feature is very generic and should work for all ELF targets.
Is there anyone who can make a decision to enable it for all ELF
targets by default?

> Also, your comment seems to imply that this feature, which is conceptually completely generic, can only ever be delivered to one target platform on one OS because there aren’t maintainers for other flavors.  That doesn’t seem right.
>

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> This feature is very generic and should work for all ELF targets.
> Is there anyone who can make a decision to enable it for all ELF
> targets by default?

Not necessarily any ONE, but the group of Global Maintainers for
the binutils project?

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Thu, Jan 8, 2015 at 7:40 PM, Joel Brobecker <brobecker@adacore.com> wrote:
>> This feature is very generic and should work for all ELF targets.
>> Is there anyone who can make a decision to enable it for all ELF
>> targets by default?
>
> Not necessarily any ONE, but the group of Global Maintainers for
> the binutils project?
>

Among global maintainers, Nick and Alan are the most active ones.  I
CCed them here.

Nick, Alan, should we enable compressed debug section in
assembler for all ELF targets?  If not all ELF targets, should we
enable it for all Linux targets?

Thanks.


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Tristan Gingold]

> On 09 Jan 2015, at 04:49, H.J. Lu <hjl.tools@gmail.com> wrote:
> 
> On Thu, Jan 8, 2015 at 7:40 PM, Joel Brobecker <brobecker@adacore.com> wrote:
>>> This feature is very generic and should work for all ELF targets.
>>> Is there anyone who can make a decision to enable it for all ELF
>>> targets by default?
>> 
>> Not necessarily any ONE, but the group of Global Maintainers for
>> the binutils project?
>> 
> 
> Among global maintainers, Nick and Alan are the most active ones.  I
> CCed them here.
> 
> Nick, Alan, should we enable compressed debug section in
> assembler for all ELF targets?  If not all ELF targets, should we
> enable it for all Linux targets?

I am not sure that enabling compression for all ELF targets is a good
idea.  There are many debuggers other than GDB that don't know yet
how to handle compressed debug section.

I would say this doesn't apply directly to Linux, but I am also sure
there are many tools using debug info that don't know how to deal
with compressed sections.

Maybe it would be simpler if the default were controlled by gcc.

Tristan.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Jan 9, 2015 at 12:32 AM, Tristan Gingold <gingold@adacore.com> wrote:
>
>> On 09 Jan 2015, at 04:49, H.J. Lu <hjl.tools@gmail.com> wrote:
>>
>> On Thu, Jan 8, 2015 at 7:40 PM, Joel Brobecker <brobecker@adacore.com> wrote:
>>>> This feature is very generic and should work for all ELF targets.
>>>> Is there anyone who can make a decision to enable it for all ELF
>>>> targets by default?
>>>
>>> Not necessarily any ONE, but the group of Global Maintainers for
>>> the binutils project?
>>>
>>
>> Among global maintainers, Nick and Alan are the most active ones.  I
>> CCed them here.
>>
>> Nick, Alan, should we enable compressed debug section in
>> assembler for all ELF targets?  If not all ELF targets, should we
>> enable it for all Linux targets?
>
> I am not sure that enabling compression for all ELF targets is a good
> idea.  There are many debuggers other than GDB that don't know yet
> how to handle compressed debug section.
>
> I would say this doesn't apply directly to Linux, but I am also sure
> there are many tools using debug info that don't know how to deal
> with compressed sections.
>
> Maybe it would be simpler if the default were controlled by gcc.
>

Most of the consumers of the outputs of assembler are binutils
tools.   Linker won't generate compressed debug sections by
default.  ld.bfd can't even generate compressed debug sections and
gold --compress-debug-sections has an issue:

https://sourceware.org/bugzilla/show_bug.cgi?id=17819

Since debuggers mostly deal with the outputs of linker,
they won't even see the compressed debug sections.  Even
if there is a problem, we can run objcopy to decompress the debug
sections.


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Alan Modra]

On Fri, Jan 09, 2015 at 05:39:29AM -0800, H.J. Lu wrote:
> Since debuggers mostly deal with the outputs of linker,
> they won't even see the compressed debug sections.

Yes, I think some of the worry up-thread was due to thinking your
patch affected linker output.  Even so, I don't think it is a good
change.  Relocatable object files are generally temporary;  You
compile package source, install the final linked binaries, then clean
up build directories.  An optimisation that reduces temporary file
size obviously helps with disk storage, and may even reduce link
time.  However there are some cases where relocatable object files are
installed, for example, crt*.o, library archives and kernel modules.
If you then try to use those object files with an older toolchain or
one built without zlib then you'll run into difficulty.

-- 
Alan Modra
Australia Development Lab, IBM

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Sun, Jan 11, 2015 at 8:29 PM, Alan Modra <amodra@gmail.com> wrote:
> On Fri, Jan 09, 2015 at 05:39:29AM -0800, H.J. Lu wrote:
>> Since debuggers mostly deal with the outputs of linker,
>> they won't even see the compressed debug sections.
>
> Yes, I think some of the worry up-thread was due to thinking your
> patch affected linker output.  Even so, I don't think it is a good
> change.  Relocatable object files are generally temporary;  You
> compile package source, install the final linked binaries, then clean
> up build directories.  An optimisation that reduces temporary file
> size obviously helps with disk storage, and may even reduce link
> time.  However there are some cases where relocatable object files are
> installed, for example, crt*.o, library archives and kernel modules.

Kernel modules don't have compressed debug sections since they
are generated by ld which doesn't output compressed debug
sections.  As for crt*.o and library archives, I would say they should
be older than your binutils.  At least, that is my case.

> If you then try to use those object files with an older toolchain or
> one built without zlib then you'll run into difficulty.

We can require zlib for targets where compression is enabled by default.

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:25 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> > If you then try to use those object files with an older toolchain or
>> > one built without zlib then you'll run into difficulty.
>>
>> We can require zlib for targets where compression is enabled by default.
>
> Given how difficult it is for me to get someone with approval
> privileges to even comment on the patch that allows someone to
> build with zlib at a non-standard location, I would say that
> this is not the friendliest option at the moment.
>

What is wrong with in-tree zlib?


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> > If you then try to use those object files with an older toolchain or
> > one built without zlib then you'll run into difficulty.
> 
> We can require zlib for targets where compression is enabled by default.

Given how difficult it is for me to get someone with approval
privileges to even comment on the patch that allows someone to
build with zlib at a non-standard location, I would say that
this is not the friendliest option at the moment.

I know the patch I proposed is not perfect, but I believe it is
the most pragmatic under the circumstances. I would also accept
and understand a reviewer's rejection of my patch; but since
the alternatives are going to be a significant amount of work
for reasons I do not believe are sufficiently important, I will
pass on that.

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> > Given how difficult it is for me to get someone with approval
> > privileges to even comment on the patch that allows someone to
> > build with zlib at a non-standard location, I would say that
> > this is not the friendliest option at the moment.
> 
> What is wrong with in-tree zlib?

Can you explain further?

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:42 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Feb 20, 2015 at 4:38 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>>> > Given how difficult it is for me to get someone with approval
>>> > privileges to even comment on the patch that allows someone to
>>> > build with zlib at a non-standard location, I would say that
>>> > this is not the friendliest option at the moment.
>>>
>>> What is wrong with in-tree zlib?
>>
>> Can you explain further?
>>
>
> Sorry I didn't realize that we only put zlib in gcc tree, not in
> binutils+gdb tree.

BTW, that may be one reason why GCC developers, like me,
got confused by your zlib patch.

> Will copy it over from gcc tree work?
>
> --
> H.J.



-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:38 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> > Given how difficult it is for me to get someone with approval
>> > privileges to even comment on the patch that allows someone to
>> > build with zlib at a non-standard location, I would say that
>> > this is not the friendliest option at the moment.
>>
>> What is wrong with in-tree zlib?
>
> Can you explain further?
>

Sorry I didn't realize that we only put zlib in gcc tree, not in
binutils+gdb tree.
Will copy it over from gcc tree work?

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:38 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> > Given how difficult it is for me to get someone with approval
>> > privileges to even comment on the patch that allows someone to
>> > build with zlib at a non-standard location, I would say that
>> > this is not the friendliest option at the moment.
>>
>> What is wrong with in-tree zlib?
>
> Can you explain further?
>

Sorry I didn't realize that we only put zlib in gcc tree, not in
binutils+gdb tree.
Will copy it over from gcc tree work?

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:38 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> > Given how difficult it is for me to get someone with approval
>> > privileges to even comment on the patch that allows someone to
>> > build with zlib at a non-standard location, I would say that
>> > this is not the friendliest option at the moment.
>>
>> What is wrong with in-tree zlib?
>
> Can you explain further?
>

Sorry I didn't realize that we only put zlib in gcc tree, not in
binutils+gdb tree.
Will copy it over from gcc tree work?

-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Fri, Feb 20, 2015 at 4:42 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Fri, Feb 20, 2015 at 4:38 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>>> > Given how difficult it is for me to get someone with approval
>>> > privileges to even comment on the patch that allows someone to
>>> > build with zlib at a non-standard location, I would say that
>>> > this is not the friendliest option at the moment.
>>>
>>> What is wrong with in-tree zlib?
>>
>> Can you explain further?
>>
>
> Sorry I didn't realize that we only put zlib in gcc tree, not in
> binutils+gdb tree.

BTW, that may be one reason why GCC developers, like me,
got confused by your zlib patch.

> Will copy it over from gcc tree work?
>
> --
> H.J.



-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> Sorry I didn't realize that we only put zlib in gcc tree, not in
> binutils+gdb tree.
> Will copy it over from gcc tree work?

I think it would. But distro vendors might want to have an option
to be able to build with the system zlib.

-- 
Joel

Re: PATCH: Compress debug sections for Linux/x86 by default

[H.J. Lu]

On Mon, Feb 23, 2015 at 11:16 PM, Joel Brobecker <brobecker@adacore.com> wrote:
>> Sorry I didn't realize that we only put zlib in gcc tree, not in
>> binutils+gdb tree.
>> Will copy it over from gcc tree work?
>
> I think it would. But distro vendors might want to have an option
> to be able to build with the system zlib.
>

That is why --with-system-zlib is here.


-- 
H.J.

Re: PATCH: Compress debug sections for Linux/x86 by default

[Joel Brobecker]

> >> Sorry I didn't realize that we only put zlib in gcc tree, not in
> >> binutils+gdb tree.
> >> Will copy it over from gcc tree work?
> >
> > I think it would. But distro vendors might want to have an option
> > to be able to build with the system zlib.
>
> That is why --with-system-zlib is here.

At first sight, and at the technical level, this seem to provide
the tools needed to support builds with zlib.

-- 
Joel