Re: Improvements to the gcc completion script

[Jacob Gelbman <gelbman@xxxxxxxxx>]

> On Mar 18, 2021, at 11:51 PM, Jun T <takimoto-j@xxxxxxxxxxxxxxxxx> wrote:
> 
> Thank you for lots of improvements!
> 
> Please consider the following points (some of them already existed
> before your improvements):
> 
> 
> In args for clang (line 509 and below), there are several duplicated options
> such as -fshow-source-location, -lazy_framework, etc.

Thanks, not sure how I completely missed some of these duplicates. -fshow-source-location, -lazy_framework, -fcaret-diagnostics, -fdiagnostics-fixit-info, -fdiagnostics-print-source-range-info, -init, -install_name, -interface-stub-version, -keep_private_externs, -lazy_library.

> 
> Some options (such as -fauto-profile) can take optional argument.
> For these, it would be better to use
> --fauto-profile=-::message:action
> instead of duplicating it with and without arg.
> 

I combined these options into one option with an optional argument: -fauto-profile, -fcf-protection, -fmemory-profile, -fprofile-instr-generate, -fprofile-instr-use, -fprofile-sample-use, -fsanitize-memory-track-origins, -fsanitize-recover, -fsanitize-trap, -fsave-optimization-record, -gz

I made sure -ftemplate-depth takes the extra - at the end, this option should look like -ftemplate-depth-5 not -ftemplate-depth5.

> 
> I think LD_LIBRARY_PATH is not used at link time (it is used only at run time),
> and we can ignore it when completing libraries ( -l<TAB> ). Is this correct?

I thought the same but wasn't 100% sure, so I left it in, but I tried it and it doesn't work, so I removed using this variable.

> 
> TODO (you can left this for someone else):
> Add directories specified by -L/path/to/dir on the command line to library_path.

I did it.

> 
> 
>> Added the -framework option, and completed the names of frameworks it can take. frameworks are macOS's style of library.
> 
> -F, -iframework and -iframeworkwithsysroot can be specified multiple times.
> Please use '*-F' etc for optspec.
> For -iframework you can use '_files -/' as action (like -F).

Okay.

> # -iframeworkwithsysroot is meaningful only with -isysroot, and, ideally,
> # '-iframeworkwithsysroot <TAB>' should offer directory relative to isysroot.
> # But I think such an elaboration would not be necessary.
> 
> -framework, -lazy_framework and -weak_framework can also be specified multiple times.
> The latter two can use '->framework' as action (as -framework).

Okay.

> 
> It would be better to include Framework-related stuff only if '$OSTYPE = darwin*'.

I think clang on linux would recognize the -framework option and pass it on to the linker, which would probably fail. but, I think there's a way of setting up linux to be able to compile Objective-C programs with frameworks using gnustep or some other collection. I've never done it though.

> 
> I don't have any detailed knowledge of how Apple's command line tools work,
> but probably it would better to use the output of 'xcrun --show-sdk-path' instead of 
> /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk.

I would use use clang -Wl,-v but it could be a bit tricky to get it to cache correctly, so I left that one be for now.
> 
> TODO: Add directory specified by -F/path/to/dir to framework_path

I did this.

> 
>> I changed the input file type because sometimes it's useful to include library files on the command line, and these have extensions .so*, .a, .dylib, or even with no extension like in mac's frameworks. I thought it better if it just completed all files and not just .c, .cpp, .m, etc.
> 
> I believe most users will not like this. Instead, please use something like 
> zstyle ':completion:*:*:gcc:argument-rest:*' file-patterns '*'
> for your personal customization.

I put it back to the way it was, but my vote is to change it.

> 
> 
>> I removed the matchspec "r:|[_-]=*" from the _arguments call because there were certain options where I would try to complete them, and they positioned the cursor elsewhere unexpectedly. for example, I wrote "clang --opt<tab>" and after it completed up until --optimiz, it positioned me on the second - when I expected to be at the end of the fully completed option, --optimize=. also it didn't fully complete, it left it at --optimiz.
> 
> I can't reproduce your problem.
> The matcher 'r:|[_-]=*' is quite useful and better not to be removed since it allows
> gcc -fd-s-l<TAB>
> to be completed to
> gcc -fdiagnostics-show-location=
> 

I miswrote how to reproduce it. You type --opti<tab>, not --opt<tab>, then you'll see it position the cursor -<here>-optimiz. This sort of thing is especially annoying if you bind tab to expand-or-complete-prefix.

> 
>> I added a completion for arguments to -Werror= ... (snip)
> 
> Thanks. I think -Werror can also be specified multiple times ('*-Werror=-').

I fixed it.

> 
> ----
> Jun