Re: [Xen-devel] [PATCH] configure: Check for flex

[Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>]

Roger Pau Monne writes ("Re: [Xen-devel] [PATCH] configure: Check for flex"):
> El 15/04/2012, a las 19:43, Jean Guyader escribió:
> > libxl require the command flex to be present.
> > Verify in the configure script that the flex
> > command exsits.
> > 
> > Signed-off-by: Jean Guyader <jean.guyader@xxxxxxxxx>
> 
> 
> I've already sent a patch for this, detecting and setting Flex and Bison at 
> configure, and printing a pretty error message if libxl needs them and they 
> are not found:
> 
> http://lists.xen.org/archives/html/xen-devel/2012-04/msg00923.html

(I'm afraid that patch is still in my (enormous) backlog, but:)

I'm not very convinced that that patch is an improvement.  All it
does, effectively, is change the error message from "flex: not found"
to a custom one which effectively says "I couldn't find flex".

If flex is not available, and the timestamps indicate the file needs
to be rebuilt, we have two choices, corresponding to two possible
situations:
  1. Assume that the problem is simply timestamp skew, and allow
     the build to continue without regenerating the file (although
     we should probably print a warning)
  2. Assume that the user has edited (or patched) the flex source
     code, and stop with an error

Of these I think 1. is preferable.  In the latter case, if the user
edited it themselves they will hopefully be reading the make output
and see the warning; whereas if the user applied a patch, the patch
should update the flex output too.

In practice we update these files rarely and of course we always
commit a corresponding change.  So doing 1. will only adversely affect
a small minority of developers.  Whereas doing 2. seems to cause
regular annoyance to many people who don't necessarily know what's
going on.

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel