Re: [PATCH] yacc: extend and improve tests

[Ralf Wildenhues]

* Stefano Lattarini wrote on Sat, Jan 08, 2011 at 08:33:28PM CET:
> On Saturday 08 January 2011, Ralf Wildenhues wrote:

> > > --- /dev/null
> > > +++ b/tests/yacc-d-basic.test
> > > @@ -0,0 +1,157 @@
> > > +#! /bin/sh
> > > +# Copyright (C) 2011 Free Software Foundation, Inc.
> > 
> > If this supersedes yacc3.test, then does it not also derive from it?
> >
> Actually no; I've basically written it from scratch (with some help
> from `yacc-basic.test'). 

Ignore my comment then.

> > > +# Even the generated header file is renamed when target-specific YFLAGS
> > > +# are used.  This might not be the best semantic, but it has been in 
> > > place
> > 
> > semantics
> > 
> I went for "behaviour" instead.

Not to get into a British vs. US English bikeshed argumentation, but I
usually use the latter, just because it's more prevalent in the code
already, and I mildly prefer consistency.  So if you don't mind, use
"behavior", but otherwise I'll try not to make this a topic any further.

> > > +# for quite a long time, so just go along with it for now.
> > 
> > Just out of curiosity: can you explain why this may not be good
> > semantics?
> >
> Because the details of outfile renamings when per-object flags are
> involved should be (as much as possible) an internal detail of
> automake.  But in this case this detail must be exposed, since the
> output header file might be #included in other files (such as in
> the examples in this same testcase).  Not a big deal, but noting
> this wart in a comment couldn't hurt either IMHO.

Sure.  I really was interested only.

> > > +$MAKE -s echo-distcom | grep '[ /]zardoz-parse.c '
> > > +$MAKE -s echo-distcom | grep '[ /]zardoz-parse.h '
> > 
> > Did you try with some non-GNU make (e.g., heirloom or BSD make)?
> >
> Yes; NetBSD make (debian port) and Heirloom/Solaris make works
> correctly, while FreeBSD make fails the distcheck with:
>   ...
>   ERROR: files left in build directory after distclean:
>   ./foo/parse.h
>   ./foo/parse.c
>   ./bar/parse.h
>   ./bar/parse.c
>   ./baz/zardoz-parse.h
>   ./baz/zardoz-parse.c
>   *** Error code 1
> 
> There are failures in other tests with FreeBSD (and Solaris) make,
> but I'd rather leave them in place as they expose real issues (no
> point in sweeping the dirt under the rug to have a 100% testsuite
> pass in any case, right?).

ACK.


> > > +# Sanity check on distribution.
> > 
> > s/on //
> >
> Really? Sounds weird to me (but I made the edit anyway).

Oops.  Hmm.  I guess that depends on whether you read 'check' as noun or
verb.  If the latter, then probably s/on/the/ would've been better
though.

> I will push soonish (tonight or tomorrow).

Thanks!
Ralf