[Pkg-pascal-devel] castle-game-engine test fails on i386

[Paul Gevers]

Hi Michalis,

On 06-01-15 05:07, Michalis Kamburelis wrote:
> That's weird --- the testsuite is checked on Linux i386 (Debian
> stable) every night, as part of the nightly build on
> http://michalis.ii.uni.wroc.pl/castle-engine-snapshots/ . It was Ok at
> engine 5.1.1 release. It also works fine now, also on my system at
> home, with engine SVN (on Debian testing, also on i386).

I see.

> Basically, you can debug such assertion failure as any other
> exception.

Problem is, I have zero experience doing that in Pascal (yes, weird for
a Pascal packager, but true).

> I also committed now a small improvement to the testsuite,
> it uses new AssertXxx methods that produce detailed output about
> expected vs actual value when test failed --- this will provide much
> more useful info in case of failure. See
> https://sourceforge.net/p/castle-engine/code/14330/ . This may help :)

Well, it does, but it doesn't ;):

List of failures:
  Failure:
    Message:           TTestCastleVectors.TestMatrixMultiplication:
Matrices (TMatrix4Single) are not equal:
  Expected:
    [ 0.48389998078346252 -0.89060002565383911 -0.25519999861717224
4.0651092576930199E-34 ]
    | 1.1405999660491943 0.84759998321533203 0.75919997692108154
5.8406861277703926E-34 |
    |  Nan  Nan  Nan  Nan |
    [ 0.31000000238418579 0.25999999046325684 0.92000001668930054
-4.2477981261616539E33 ]
  Actual:
    [ 0.48389998078346252 -0.89060002565383911 -0.25519999861717224
4.0651092576930199E-34 ]
    | 1.1405999660491943 0.84759998321533203 0.75919997692108154
5.8406861277703926E-34 |
    |  Nan  Nan  Nan  Nan |
    [ 0.31000000238418579 0.25999999046325684 0.92000001668930054
-4.2477981261616539E33 ]
    Exception class:   EAssertionFailedError
    Exception message: Matrices (TMatrix4Single) are not equal:
  Expected:
    [ 0.48389998078346252 -0.89060002565383911 -0.25519999861717224
4.0651092576930199E-34 ]
    | 1.1405999660491943 0.84759998321533203 0.75919997692108154
5.8406861277703926E-34 |
    |  Nan  Nan  Nan  Nan |
    [ 0.31000000238418579 0.25999999046325684 0.92000001668930054
-4.2477981261616539E33 ]
  Actual:
    [ 0.48389998078346252 -0.89060002565383911 -0.25519999861717224
4.0651092576930199E-34 ]
    | 1.1405999660491943 0.84759998321533203 0.75919997692108154
5.8406861277703926E-34 |
    |  Nan  Nan  Nan  Nan |
    [ 0.31000000238418579 0.25999999046325684 0.92000001668930054
-4.2477981261616539E33 ]

To me they look the same, but maybe NaN is not equal (I know Matlab
well, and NaN is not equal to NaN.

> If you can still reproduce the bug, can you write what FPC version do
> you use,

The current version in sid/jessie
Free Pascal Compiler version 2.6.4+dfsg-4 [2014/10/14] for i386

> and how exactly are you compiling the testsuite? I'm simply
> compiling + running like this:

You can easily check everything in the pkg-pascal archive:
http://anonscm.debian.org/cgit/pkg-pascal/castle-game-engine.git/tree/debian/rules

> cd ..../castle_game_engine/
> make clean
> cd tests
> ./compile_console.sh
> ./test_castle_game_engine -a

We do:
	@echo "--- Building"
	dh_testdir
	# Building Pascal style font file first while making sure that
	# no unit or object files polute the source tree
	${MKDIR} ${TMP_DIR}
	${FPC} -FE${TMP_DIR}  @castle-fpc.cfg examples/fonts/texturefont2pascal.lpr
	cd src/fonts ; PATH=${PATH}:${TMP_DIR} ./mk_fonts.sh

	${MAKE} all

ifeq (,$(filter nocheck,$(DEB_BUILD_OPTIONS)))
	# Code to run the package test suite.
	# Command basically copied from tests/compile_console.sh while
	# making sure that no unit or object files polute the source tree
	${FPC}  -gh -dDEBUG -dTEXT_RUNNER -dNO_WINDOW_SYSTEM "$@" \
		-Fuexamples/shadow_fields/ -FE${TMP_DIR} \
		@castle-fpc.cfg tests/test_castle_game_engine.lpr
	# cd is needed to find the data files
	cd tests ; ${TMP_DIR}/test_castle_game_engine -a
endif

Do you see anything wrong here?

> Works fine with FPC 2.6.2, 2.6.4 and fairly recent FPC from SVN
> (2.7.1). I tested with FPC from freepascal.org (not the Debian
> packaged one).

So seeing that the content is visually the same, I would like to see the
content when YOU run the test.

Paul


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/pkg-pascal-devel/attachments/20150106/a4322ebd/attachment.sig>