remove _RWSTD_NO_MEMBER_TEMPLATES and workarounds

• Key: STDCXX-978
• Type: Task
• Status: In Progress
• Priority: Major
• Assignee: Eric Lemings
• Reporter: Martin Sebor
• Component/s: Configuration
• Affects Version/s: 4.3.x
• Fix Version/s: 4.3.0

Description

Quoting from the following discussion:

Martin Sebor wrote:
> Travis Vitek wrote:
>>
>>
>>> Eric Lemings wrote:
>>>
>>>> Travis Vitek wrote:
>>>>
>>>>> Eric Lemings wrote:
>>>>>
>>>>> How about member templates? Are these unilaterally supported by all
>>>>> compilers now?
>>>>>
>>>> From below...
>>>>
>>>> _RWSTD_NO_INLINE_MEMBER_TEMPLATES /* not used at all */
>>> Well there's an INLINE_MEMBER_TEMPLATES.cpp check and then there's a
>>> separate MEMBER_TEMPLATES.cpp check. I assume they check different
>>> things.
>>
>> Yes, I realize this. The former checks that inline member templates are
>> supported. The latter checks that member templates can be defined
>> outside the body of the class.
>>
>> AFAICT, inline member templates are used in library code without guards,
>> so you can safely assume support for 4.2.x and later.
>>
>> If you want anything further than that [i.e. member templates defined
>> outside the body of the class, member template overloads, ...], you have
>> to look to see what all of the supported compilers allow. If one or more
>> of the currently supported compilers defines _RWSTD_NO_MEMBER_TEMPLATES,
>> then you will likely have to work around the issue in any new code. If
>> none of them define it, then some decision will have to be made if it is
>> safe to assume support for other compilers or not.
>
> Starting with 4.3.0, I think we can safely assume full support for
> member templates and remove all workarounds involving the macros
> as well as the macros (and the config test) themselves. The only
> thing I'd double-check before doing this cleanup is string when
> using HP aCC 3 because of an extern template bug.
>
> I don't even think there is a compiler that we target with 4.2.x
> that doesn't support member templates in any form. IIRC, the last
> one that had trouble with member templates (their out-of-line
> definitions) was MSVC 6.
>
> Martin

Comments

[Martin Sebor]

Assigned to Brad.

[Eric Lemings]

I assume this issue also applies to _RWSTD_NO_MEMBER_TEMPLATE_OVERLOAD since it is only referenced in the <bitset> header.

[Eric Lemings]

Also I can find no config test or other conditions for which _RWSTD_NO_TEMPLATE_ON_RETURN_TYPE is defined.

[Martin Sebor]

This issue was meant to apply only to _RWSTD_NO_MEMBER_TEMPLATES. The _RWSTD_NO_MEMBER_TEMPLATE_OVERLOAD macro is #defined for a different limitation/bug. That being said, if none of the compilers on our test matrix #defines the macro feel free to remove it as well. To check, grep people.apache.org://www/stdcxx.dev.apache.org/builds/4.3.x/logs/*-cfg-h.gz.txt for the string "^ *# *define *_RWSTD_NO_MEMBER_TEMPLATE_OVERLOAD *"

The config test for _RWSTD_NO_TEMPLATE_ON_RETURN_TYPE is TEMPLATE_ON_RETURN_TYPE.cpp.

[Eric Lemings]

Checked all log files in the {/www/stdcxx.dev.apache.org/builds/4.3.x/logs} directory. None of the macros {_RWSTD_NO_MEMBER_TEMPLATES}, {_RWSTD_NO_MEMBER_TEMPLATE_OVERLOAD}, and {_RWSTD_NO_TEMPLATE_ON_RETURN_TYPE} are being defined. I'm leaving the return type macro untouched however since it is unrelated to this issue (unlike the overload macro).

[Martin Sebor]

This will not affect the 4.3.0 release but it does (or did until r678913) affect the 4.3.x branch, so replaced 4.3.0 with 4.3.x in Affects Version/s.