[Lldb-commits] [PATCH] D17724: Replace getopt with llvm::cl in lldb driver

[Pavel Labath via lldb-commits]

labath added a comment.

In http://reviews.llvm.org/D17724#364450, @clayborg wrote:

> As long as both long and short options are still supported? Can you still type any of:
>
> - --arch=x86_64
> - --arch x86_64
> - -arch=x86_64
> - -arch x86_64
> - -ax86_64
> - -a x86_64


All the variants work (including -a=x86_64) **except** the "-ax86_64" version. Not sure if it's that important to maintain this one.

> Are we able to track which options can be used with other options with the llvm solution? I didn't look too close. There were bits in the old option definitions which defined with options could be specified with which other options.


These bits were there, but they weren't used for any verification purposes, only for formatting the `--help` output, which I did not find altogether useful anyway, as it has too many options and variants (some of which are wrong).

  lldb -h
  lldb -v [[--] <PROGRAM-ARG-1> [<PROGRAM_ARG-2> ...]]
  lldb -a <arch> -f <filename> [-c <filename>] [-s <filename>] [-o <none>] [-S <filename>] [-O <none>] [-k <none>] [-K <filename>] [-Q] [-b] [-e] [-x] [-X] [-l <script-language>] [-d] [[--] <PROGRAM-ARG-1> [<PROGRAM_ARG-2> ...]]
  lldb -n <process-name> -w [-s <filename>] [-o <none>] [-S <filename>] [-O <none>] [-k <none>] [-K <filename>] [-Q] [-b] [-e] [-x] [-X] [-l <script-language>] [-d]
  lldb -p <pid> [-s <filename>] [-o <none>] [-S <filename>] [-O <none>] [-k <none>] [-K <filename>] [-Q] [-b] [-e] [-x] [-X] [-l <script-language>] [-d]
  lldb -P
  lldb -r [<none>] -R <none>

We can't get llvm to produce this kind of output for us, but we can use "option categories". By default, we have "general" and "generic" categories ( :P ), but I was thinking about adding a "repl" category (for --repl, --repl-language) and a "scripting category" (all --source and --one-line options, --batch, etc.). IMHO, this would help the output readability more than the previous "option sets" variant.

This would not help the enforcement of option combinations in a declarative manner, but as we have introspection, it wouldn't be too hard to express "no option from the 'repl' category can be used with other options" with a bit of code.

> The other thing that might throw people for a loop is if llvm doesn't support this style:

> 

> % lldb /bin/ls --arch=x86_64 -- -lAF


This works as expected.

So, yes, there are tiny regressions, but these are not important IMO, and I think this makes a nice code simplification. However, if I'm not interested in shoving this in if people don't approve, so... let me know what you think


http://reviews.llvm.org/D17724