Build flag changes [ was Re: Large patch to merge in changes for gcc 4.3 ]

Build flag changes [ was Re: Large patch to merge in changes for gcc 4.3 ]

[John Dallaway]

Jifl

[ moving to the ecos-maintainers list ]

Jonathan Larmour wrote:

> Therefore if we're going to be touching all HAL's CYGBLD_GLOBAL_CFLAGS,
> it's an opportunity to centralise these warning flags, which are entirely
> generic and are distributed identically across all HALs. I propose adding a
> CYGBLD_GLOBAL_WARNFLAGS CDL option in hal/common. And I will remove all the
> standard warning flags[1] from all HALs, and replace it with a reference to
> that option. More specifically I suggest that CYGBLD_GLOBAL_WARNFLAGS is an
> option which consists of a concatenation of CYGBLD_GLOBAL_WARNFLAGS_C,
> CYGBLD_GLOBAL_WARNFLAGS_CXX and CYGBLD_GLOBAL_WARNFLAGS_COMMON. Each of
> these takes their default_value from an identically named option with a
> _DEFAULT suffix. This allows package CDL to control the default, as well as
> letting users override it. It also starts to give us a route to avoid the
> horrible hackery in pkgconf/rules.mak to eliminate language-specific flags
> by replacing it with something properly controlled. If anyone has a better
> proposal, I'd like to hear it, although be warned that if it is more
> effort, then they can consider themselves automatically volunteered to do it!

I agree that the rules.mak hackery should be eliminated but your
proposal appears to involve a lot of CDL editing and suggests a need for
updated host tools to read and process the new CYGBLD_GLOBAL_WARN*
options. I'm concerned about the time overhead for debugging and testing
such changes at a late stage in the release cycle. Could your proposal
be split into phases to keep changes to a minimum for eCos 3.0 while
still avoiding multiple edits to the platform HAL CDL scripts?

John Dallaway

Re: Build flag changes [ was Re: Large patch to merge in changes for gcc 4.3 ]

[Jonathan Larmour]

John Dallaway wrote:
> Jonathan Larmour wrote:
> 
>> Therefore if we're going to be touching all HAL's CYGBLD_GLOBAL_CFLAGS,
>> it's an opportunity to centralise these warning flags, which are entirely
>> generic and are distributed identically across all HALs. I propose adding a
>> CYGBLD_GLOBAL_WARNFLAGS CDL option in hal/common. And I will remove all the
>> standard warning flags[1] from all HALs, and replace it with a reference to
>> that option. More specifically I suggest that CYGBLD_GLOBAL_WARNFLAGS is an
>> option which consists of a concatenation of CYGBLD_GLOBAL_WARNFLAGS_C,
>> CYGBLD_GLOBAL_WARNFLAGS_CXX and CYGBLD_GLOBAL_WARNFLAGS_COMMON. Each of
>> these takes their default_value from an identically named option with a
>> _DEFAULT suffix. This allows package CDL to control the default, as well as
>> letting users override it. It also starts to give us a route to avoid the
>> horrible hackery in pkgconf/rules.mak to eliminate language-specific flags
>> by replacing it with something properly controlled. If anyone has a better
>> proposal, I'd like to hear it, although be warned that if it is more
>> effort, then they can consider themselves automatically volunteered to do it!
> 
> I agree that the rules.mak hackery should be eliminated but your
> proposal appears to involve a lot of CDL editing

Only by script. And this is a step towards eliminating the hackery, but
won't yet do so in isolation.

> and suggests a need for
> updated host tools to read and process the new CYGBLD_GLOBAL_WARN*
> options.

I was intending a default_value for the platform CFLAGS of the existing
options string concatenated with CYGBLD_GLOBAL_WARNFLAGS

e.g.
            default_value { "-mcpu=arm9 -Wall -Wpointer-arith
-Wstrict-prototypes -Winline -Wundef -Woverloaded-virtual -g -O2
-ffunction-sections -fdata-sections -fno-rtti -fno-exceptions" }

becomes

            default_value { "-mcpu=arm9 -g -O2 -ffunction-sections
-fdata-sections -fno-rtti -fno-exceptions " . CYGBLD_GLOBAL_WARNFLAGS }

This sort of substitution is very amenable to being done by script.

In due course I would expect we'd remove CYGBLD_GLOBAL_CFLAGS and build in
direct host tools knowledge, no doubt with the ever-distant rewritten
makefile generator (then CYGBLD_GLOBAL_WARNFLAGS_C is only used for C
files, CYGBLD_GLOBAL_WARNFLAGS_CXX is only used for C++ files, etc.) But
not for eCos 3.0.

With this change, after the release branch is taken, I can simply change
CYGBLD_GLOBAL_WARNFLAGS_COMMON_DEFAULT to add -Wno-write-strings.

> I'm concerned about the time overhead for debugging and testing
> such changes at a late stage in the release cycle. Could your proposal
> be split into phases to keep changes to a minimum for eCos 3.0 while
> still avoiding multiple edits to the platform HAL CDL scripts?

I'd want to do it by script in any case if it was for more than a handful
of targets, so I don't see any reason not to do all targets.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine