This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Recent simulator patches broke many sims


On Tuesday 26 March 2013 13:12:38 Hans-Peter Nilsson wrote:
> From: Mike Frysinger <vapier@gentoo.org>
> On Sunday 24 March 2013 19:23:28 Hans-Peter Nilsson wrote:
> > You need a board (make check RUNTESTFLAGS=--target_board=$board
> > with e.g. board=cris-sim).  All boards are in "recent"
> > dejagnu-1.5 IIRC and most in ancient dejagnu-1.4.4.  You need
> > installed binutils (e.g. in some temp location added to PATH for
> > the duration of the test-run) for each sim configuration as
> > mentioned.  I don't run with target gcc; not needed for the
> > level of smoke test I'm after and I guess not for this change
> > either.
> > 
> > that's not entirely true.  many (all but cris?)
> 
> I don't think it's different but I don't plan to test without...
> 
> > sims run &
> > pass just fine without needing to explicitly pass magic flags.
> 
> It is entirely true that when a board is specified, all work.
> 
> Now that you mention it, someone *did* do some changes to allow
> simulator tests to run without specifying a board - IIRC in some
> situations, assuming no special linker flags or such needed and
> no compiler toolchain (or no flags or libraries using simulator
> hooks).  Reading ChangeLogs it seems it was you, on 2010-04-26.

yes, that was the commit date, but i'd been using it longer than that :)

> > i know the Blackfin and frv sims can build & run pretty much
> > all their tests w/out requiring board flags.
> > 
> > imo, requiring manual board selection like this is archaic for
> > no good reason.
> 
> One good reason IMO is that when specifying a board, all
> toolchain parts test alike, rather than sim (after 2010-04-26)
> being a special case (and binutils, mostly for not needing to
> run things to avoid FAILs or hanging tests).

i see it the other way.  doing `make check` for all these packages work w/out 
the user having to dig into esoteric tools (that often times lack 
documentation, or any real semblance of an entry point).

for binutils/gcc/etc..., there are plenty of compile/link/etc... tests that 
can be done without needing to know about a board.  the sim is special in that 
testing it requires running code, so it's entirely reasonable for the default 
`make check` to setup a state where the tests actually work.

if the default `make check` for cris doesn't, then that should be fixed to 
provide a reasonable default.  if you're telling everyone "in order to test 
the cris sim, you have to use RUNTESTFLAGS=--target_board=cris-sim", then make 
that the default board already.  there's absolutely no reason to force this 
upon people who are trying to hack on the core sim and/or other sims, and want 
to verify they didn't break other sims.

for many sims, it's hard enough as it is just to try and guess the magic tuple 
that'll pass all the deprecated/enable checks.
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]