[cfe-commits] [PATCH] Add multiple prefixes to Option.

[Daniel Dunbar]

On Fri, Oct 19, 2012 at 1:08 PM, Michael Spencer <bigcheesegs at gmail.com> wrote:
> On Fri, Oct 19, 2012 at 12:48 PM, Daniel Dunbar <daniel at zuster.org> wrote:
>> On Fri, Oct 19, 2012 at 12:39 PM, Michael Spencer <bigcheesegs at gmail.com> wrote:
>>> On Fri, Oct 19, 2012 at 10:38 AM, Daniel Dunbar <daniel at zuster.org> wrote:
>>>> Hi Michael,
>>>>
>>>> I understand what this patch does, but I don't really understand the
>>>> motivation for it, yet.
>>>>
>>>> Why is this the "main feature" needed for sane gnu-ld and link command
>>>> line support?
>>>>
>>>> It seems to me, that what you really need for those is a completely
>>>> separate option parsing table. I can see how this patch could be
>>>> useful if you wanted to accept the same options for the link.exe
>>>> parsing just with a '/' at the front, but that doesn't seem right to
>>>> me. It seems to me that the link.exe parsing or cl.exe parsing would
>>>> end up using a completely separate parsing table.
>>>>
>>>> Can you explain?
>>>>
>>>>  - Daniel
>>>
>>> The option tables would be completely separate. This is needed because
>>> link supports both / and - prefixes for all options. gnu-ld supports -
>>> and -- for almost all options.
>>
>> If the option tables are separate, can't we dramatically simplify this
>> patch by making this a feature of the option table then? So that an
>> option table can opt in to explicitly specifying all prefixes, and
>> then we don't need to clutter up the Clang/GCC option table. Option
>> tables which didn't opt-in to that would be required to supply a list
>> of possible prefixes.
>>
>>  - Daniel
>
> This would work if clang and gcc used only one prefix. There are 36
> options that use both - and --, and 103 that use --.

I think you misunderstood me, I was proposing initializing the option
table with a special mode that says "all prefixes are defined as part
of the option". Only the Clang/GCC table would use that.

 - Daniel

> I can simplify this patch by defaulting to - via tblgen and only
> changing those 139 options that are special.
>
> - Michael Spencer
>
>>> From the ld man page:
>>>
>>> "For options whose names are multiple letters, either one dash or two
>>> can precede the option name; for example, -trace-symbol and
>>> --trace-symbol are equivalent. Note---there is one exception to this
>>> rule. Multiple letter options that start with a lower case 'o' can
>>> only be preceded by two dashes. This is to reduce confusion with the
>>> -o option. So for example -omagic sets the output file name to magic
>>> whereas --omagic sets the NMAGIC flag on the output."
>>>
>>> - Michael Spencer
>>>
>>>>
>>>> On Thu, Oct 18, 2012 at 3:53 PM, Michael Spencer <bigcheesegs at gmail.com> wrote:
>>>>> Hi ddunbar,
>>>>>
>>>>> This patch adds support for multiple prefixes per option. This is the main feature needed for sane gnu-ld and link command line support.
>>>>>
>>>>> Most of the ["-"]'s can be refactored to a common class in tablegen, but I have a few more widespread tablegen changes I want to make before doing all the refactoring needed.
>>>>>
>>>>> http://llvm-reviews.chandlerc.com/D69
>>>>>
>>>>> Files:
>>>>>   include/clang/Driver/Arg.h
>>>>>   include/clang/Driver/CC1AsOptions.h
>>>>>   include/clang/Driver/CC1AsOptions.td
>>>>>   include/clang/Driver/CC1Options.td
>>>>>   include/clang/Driver/OptParser.td
>>>>>   include/clang/Driver/OptTable.h
>>>>>   include/clang/Driver/Option.h
>>>>>   include/clang/Driver/Options.h
>>>>>   include/clang/Driver/Options.td
>>>>>   lib/Driver/Arg.cpp
>>>>>   lib/Driver/ArgList.cpp
>>>>>   lib/Driver/CC1AsOptions.cpp
>>>>>   lib/Driver/Driver.cpp
>>>>>   lib/Driver/DriverOptions.cpp
>>>>>   lib/Driver/OptTable.cpp
>>>>>   lib/Driver/Option.cpp
>>>>>   lib/Driver/Tools.cpp
>>>>>   lib/Frontend/CompilerInvocation.cpp
>>>>>   utils/TableGen/OptParserEmitter.cpp