[OE-core] [PATCH 2/5] devtool: add configure-help subcommand

[Paul Eggleton]

On Mon, 25 Jan 2016 22:27:27 Khem Raj wrote:
> On Mon, Jan 25, 2016 at 10:21 PM, Paul Eggleton
> 
> <paul.eggleton at linux.intel.com> wrote:
> > On Tue, 26 Jan 2016 16:17:39 Paul Eggleton wrote:
> >> Hi Khem,
> >> 
> >> On Mon, 25 Jan 2016 22:14:09 Khem Raj wrote:
> >> > > On Jan 25, 2016, at 9:53 PM, Paul Eggleton
> >> > > <paul.eggleton at linux.intel.com>
> >> > > wrote:
> >> > > 
> >> > > When you need to set EXTRA_OECONF for a recipe, you need to know what
> >> > > options the configure script actually supports; the configure script
> >> > > however is only accessible from within a devshell and (at least in
> >> > > the
> >> > > case of autotooled software fetched from an SCM repository) may not
> >> > > actually exist until do_configure has run. Thus, provide a "devtool
> >> > > configure-help" subcommand that runs the configure script for a
> >> > > recipe
> >> > > with --help and shows you the output through a pager (e.g. less),
> >> > > prefaced by a header describing the current options being specified.
> >> > > 
> >> > > There is basic support for autotools, cmake and bare configure
> >> > > scripts.
> >> > > The cmake support is a little hacky since cmake doesn't really have a
> >> > > concise help option that lists user-defined knobs (without actually
> >> > > running through the configure process), however that being a design
> >> > > feature of cmake there's not much I can think of to do about that at
> >> > > the moment.
> >> > 
> >> > this option is autotools specific. We need to convey this precisely.
> >> > may
> >> > be
> >> > involve autotools in the option name or something
> >> 
> >> Whilst not all non-autoconf configure scripts provide a --help option,
> >> some
> >> do. ffmpeg/libav is one example (and one where you really do need to see
> >> the output since there are a lot of options...)
> > 
> > I meant to add, even if configure --help doesn't work (or there is no
> > configure script at all) the header this subcommand prints will work with
> > any recipe provided that do_configure is actually implemented and
> > contains some actual commands as opposed to calls to some other function.
> 
> my worry is it being associated to do_configure, and we know that only
> a subset ( although a large one) is only using autotools but not all.
> So think about avoiding that confusion.

I'm not terribly convinced there could be much confusion. FWIW, I just looked 
through the other recipes we have in OE-Core that have a call to a configure 
script  that don't use autoconf (man, zlib, x264, v86d, ed, libpostproc); the 
configure scripts for all of these at least produce some useful output when 
called with --help. I then took a look at meta-oe (the layer), and checked the 
following recipes:

* concurrencykit - works
* ddrescue - works
* fio - works
* libvpx - works
* lzip - works
* nodejs - works
* mg - works (no options, but message is sensible)
* terminus-font - configure script is supplied without the execute bit set but
  does support --help
* iscsi-initiator-utils - configure script is in a subdir so isn't found, but
  does support --help
* netkit-rsh - works
* netkit-rpc - works
* netkit-telnet - works
* netkit-tftp - works
* netkit-rusers - works
* netkit-rwho - works
* netmap - configure script is in a subdir so isn't found, but does support 
  --help
* ntimed - configure script is dumb and takes no options at all
* tvheadend - works
* mozjs - seems to break devtool modify, but configure script does support 
  --help

So there are some minor bugs to fix, but of those recipes I tested there is a 
grand total of 1 script that doesn't support --help. Honestly I think if a 
configure script exists it's reasonable to assume that 
--help is going to print something useful; even if it doesn't there shouldn't 
be any major harm and you're still going to get a printout of the do_configure 
value in which you'll be able to see what options are currently passed.

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre