code review 1778041: more soft float support. passes a lot more tests but ha...

more soft float support. passes several basic tests
but with less precision than hardware counterparts.

fixed a number of tests to output BUG when they failed.
changed the runner to distinghuish between output
and output containing ^BUG

[kaib, 2010-07-09 12:09:20]

Hello rsc (cc: dho, golang-dev@googlegroups.com),

I'd like you to review this change.

[kaib, 2010-07-12 10:21:49]

updated, cleaner than before

[rsc1, 2010-07-12 20:31:22]

This is changing the semantics of tests, which I'd rather not do.

It is the test's responsibility to exit 0 always.
If there is a bug, it should print one line containing
the word BUG and whatever else it feels like printing.
The number of bugs printed at the end of run is the
number of BUG lines found in the collected output.

http://codereview.appspot.com/1778041/diff/12001/13002
File test/arm-pass.txt (right):

http://codereview.appspot.com/1778041/diff/12001/13002#newcode21
test/arm-pass.txt:21: # ./cmplxdivide.go	# fail
these comments are much less useful than the old ones.
say what failed.

http://codereview.appspot.com/1778041/diff/12001/13003
File test/cmplxdivide.go (right):

http://codereview.appspot.com/1778041/diff/12001/13003#newcode38
test/cmplxdivide.go:38: bad := false
I don't like this.  Generating unexpected output should be sufficient to fail.

http://codereview.appspot.com/1778041/diff/12001/13004
File test/float_lit.go (right):

http://codereview.appspot.com/1778041/diff/12001/13004#newcode12
test/float_lit.go:12: var bad bool
same

http://codereview.appspot.com/1778041/diff/12001/13005
File test/floatcmp.go (right):

http://codereview.appspot.com/1778041/diff/12001/13005#newcode91
test/floatcmp.go:91: if bad {
same

http://codereview.appspot.com/1778041/diff/12001/13007
File test/ken/cplx4.go (right):

http://codereview.appspot.com/1778041/diff/12001/13007#newcode46
test/ken/cplx4.go:46: os.Exit(1)
same

http://codereview.appspot.com/1778041/diff/12001/13008
File test/literal.go (right):

http://codereview.appspot.com/1778041/diff/12001/13008#newcode225
test/literal.go:225: os.Exit(1)
same

http://codereview.appspot.com/1778041/diff/12001/13011
File test/zerodivide.go (right):

http://codereview.appspot.com/1778041/diff/12001/13011#newcode141
test/zerodivide.go:141: bad := false
same

[kaib, 2010-07-19 09:30:44]

as per offline discussion changed to outputting bug instead of returning !0.

just to summarize, the intent was never to change the semantics of the tests.
however, some test were failing and returning 0 without any BUG output and that
made it hard to diagnose them.

http://codereview.appspot.com/1778041/diff/12001/13002
File test/arm-pass.txt (right):

http://codereview.appspot.com/1778041/diff/12001/13002#newcode21
test/arm-pass.txt:21: # ./cmplxdivide.go	# fail
On 2010/07/12 20:31:22, rsc1 wrote:
> these comments are much less useful than the old ones.
> say what failed.

these comments are auto-generated which makes them a lot more useful when
actually working on fixing things. i also found that the information didn't
actually help drive debug decisions as much as i hoped.

http://codereview.appspot.com/1778041/diff/12001/13003
File test/cmplxdivide.go (right):

http://codereview.appspot.com/1778041/diff/12001/13003#newcode38
test/cmplxdivide.go:38: bad := false
On 2010/07/12 20:31:22, rsc1 wrote:
> I don't like this.  Generating unexpected output should be sufficient to fail.
> 

Done.

http://codereview.appspot.com/1778041/diff/12001/13004
File test/float_lit.go (right):

http://codereview.appspot.com/1778041/diff/12001/13004#newcode12
test/float_lit.go:12: var bad bool
On 2010/07/12 20:31:22, rsc1 wrote:
> same
> 

Done.

[rsc1, 2010-07-19 17:03:11]

LGTM

Some of the tests were commented out because
they were painfully slow.  Are they still slow?

[kaib, 2010-07-20 12:51:45]

On 2010/07/19 17:03:11, rsc1 wrote:
> LGTM
> 
> Some of the tests were commented out because
> they were painfully slow.  Are they still slow?

On a nexus one:

export GOGC=off; time ./run-arm
real	2m46.878s

export GOGC=; time ./run-arm
real	3m9.506s

It's pretty decent when a full ./run on my mac takes ~40s.

[kaib, 2010-07-20 12:54:05]

*** Submitted as http://code.google.com/p/go/source/detail?r=e0b81ae5fb70 ***

more soft float support. passes several basic tests
but with less precision than hardware counterparts.

fixed a number of tests to output BUG when they failed.
changed the runner to distinghuish between output
and output containing ^BUG

R=rsc
CC=dho, golang-dev
http://codereview.appspot.com/1778041