Re: [Libguestfs] [PATCH] add enable_appliance flag to specify the appliance build

On Fri, Apr 28, 2017 at 05:03:58PM +0300, Pavel Butsykin wrote: > On 27.04.2017 18:28, Richard W.M. Jones wrote: >> >> On Thu, Apr 27, 2017 at 03:55:10PM +0300, Pavel Butsykin wrote: >>> The flag determines the behavior of working with the appliance, and specifically >>> the choice to use the build supermin appliance or the fixed appliance. If the >>> flag is false then the libguestfs will only use the fixed appliance, otherwise >>> the build supermin appliance. >>> >>> This patch is aimed at solving the problem with undefined behavior when >>> searching for the appliance. The libguestfs configured with --enable-appliance >>> will only use the build supermin appliance, with --disable-appliance only use >>> the fixed appliance. >>> >> >> It has to be said I like this approach even less :-/ >> > > Any arguments? What exactly is wrong? Specifically it exposes a whole new API which just serves to adjust some internal issue in libguestfs (which as I said in the other reply is really a bug in libguestfs). Adding new APIs has a high bar because we have to support them forever.

I really think the solution here is to fix the path search so it works like you'd expect a path search to work, ie: LIBGUESTFS_PATH=/a:/b:/c means look in /a for any appliance, then look in /b, then look in /c. (Whereas currently it means something like "go round and round in circles looking for different sorts of appliance".) Then change the distro so it doesn't install two appliances into a single location, which only ever worked by random chance (if indeed it ever did work).

Rich.