Re: [Patch V2] BaseTools: support the NOOPT target with the GCC tool chains

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 5 October 2016 at 15:48, Laszlo Ersek <lersek@redhat.com> wrote:
> On 10/05/16 03:30, Yonghong Zhu wrote:
>> Update the tools_def.template to add NOOPT support with GCC tool chains.
>>
>> Cc: Liming Gao <liming.gao@intel.com>
>> Cc: Laszlo Ersek <lersek@redhat.com>
>> Contributed-under: TianoCore Contribution Agreement 1.0
>> Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
>> ---
>>  BaseTools/Conf/tools_def.template | 28 ++++++++++++++++++++++++++++
>>  1 file changed, 28 insertions(+)
>
> I thought I understood what was going on, but apparently I was wrong
> about that.
>
> In this patch, we add or modify:
> - NOOPT_*_*_OBJCOPY_ADDDEBUGFLAG -- okay
> - NOOPT_GCC*_(IA32|X64|ARM|AARCH64)_CC_FLAGS -- okay
>
> So that part is fine with me. But then we also add / modify:
> - NOOPT_GCC(49|5)_AARCH64_DLINK_(FLAGS|XIPFLAGS)
> - NOOPT_GCC5_ARM_DLINK_FLAGS
>
> First I thought the latter set of changes was unnecessary, because "ld"
> didn't use "-O". I checked the manual, and I was wrong: "ld" does know /
> use "-O". So those changes are fine, I guess.
>

Yes, especially under LTO, in which case code generation is performed
during the link stage, which should adhere to the same rules as the
compiler. This not only applies to -O, but also to things like
-march/-mcpu and -mstrict-align. This is why we pass all CFLAGS to the
linker for the GCC5 LTO builds.

> But then: is the patch *complete*? Because I can see some more DLINK
> stuff, for IA32 and X64 (not just ARM and AARCH64). Is it okay to ignore
> those? For example:
>
> *_GCC5_IA32_DLINK_FLAGS          = DEF(GCC5_IA32_X64_DLINK_FLAGS) -Os
> -Wl,-m,elf_i386,--oformat=elf32-i386
>
>
> *_GCC5_X64_DLINK_FLAGS           = DEF(GCC5_X64_DLINK_FLAGS) -Os
>
> Where GCC5_X64_DLINK_FLAGS and GCC5_IA32_X64_DLINK_FLAGS even include
> -flto. (I don't know if "-flto" hampers source level debugging or not.)
>

The GCC man page documents -flto as being a bad idea, i.e.,

"""
Link-time optimization does not work well with generation of debugging
information.  Combining -flto with -g is currently experimental and
expected to produce unexpected results.
"""

(which raises a philosophical question as well, i.e., to which extent
expected unexpected results are still unexpected results. But I
digress ...)

Another note: the DEBUG build for ARM and AARCH64 is essentially NOOPT
already, not DEBUG. How does this patch intend to deal with that?