Add S390X architecture as a Travis job.

[Girgias]

This gives us a way to compile and test a big endian architecture.

[Girgias]

=====================================================================
FAILED TEST SUMMARY
---------------------------------------------------------------------
Bug #79046: NaN to int cast undefined behavior in exif [ext/exif/tests/bug79046.phpt]
Overflow in float to int cast [ext/exif/tests/float_cast_overflow.phpt]
FFI 040: Support for scalar types [ext/ffi/tests/040.phpt]
Test ResourceBundle array access and count - existing/missing keys [ext/intl/tests/resourcebundle_arrayaccess.phpt]
Test ResourceBundle implements Countable [ext/intl/tests/resourcebundle_countable.phpt]
Test ResourceBundle::__construct() - existing/missing bundles/locales [ext/intl/tests/resourcebundle_create.phpt]
Test ResourceBundle::get() and length() - existing/missing keys [ext/intl/tests/resourcebundle_individual.phpt]
Test ResourceBundle iterator [ext/intl/tests/resourcebundle_iterator.phpt]
Test ResourceBundle::getLocales [ext/intl/tests/resourcebundle_locales.phpt]
Bug #55610: ResourceBundle does not implement Traversable [ext/intl/tests/resourcebundle_traversable.phpt]
function test: nested selects (cursors) [ext/mysqli/tests/067.phpt]
Bug #36745 (LOAD DATA LOCAL INFILE doesn't return correct error message) [ext/mysqli/tests/bug36745.phpt]
mysqli_get_client_stats() [ext/mysqli/tests/mysqli_get_client_stats.phpt]
mysqli_stmt_attr_set() - mysqlnd does not check for invalid codes [ext/mysqli/tests/mysqli_stmt_attr_set.phpt]
MySQL PDO->__construct(), options [ext/pdo_mysql/tests/pdo_mysql___construct_options.phpt]
enable local infile [ext/pdo_mysql/tests/pdo_mysql_local_infile_set_on.phpt]
=====================================================================

The FFI one has been reported already at: https://bugs.php.net/bug.php?id=79472

Intl failures are due to the fact that the constuctor is failing e.g:

========DIFF========
001+ Fatal error: Uncaught IntlException: Constructor failed in /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php:4
001- bool(true)
002+ Stack trace:
003+ #0 /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php(4): ResourceBundle->__construct('es', '/home/travis/bu...')
004+ #1 {main}
005+   thrown in /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php on line 4
========DONE========

The EXIF ones seems to related to overflow behaviour (and some cast with NaN which seem highly surprising)

========DIFF========
020+   float(8.914360180802324E-41)
020-   float(NAN)
024+     float(8.914360180802324E-41)
024-     float(NAN)
========DONE========
FAIL Bug #79046: NaN to int cast undefined behavior in exif [ext/exif/tests/bug79046.phpt] 

========DIFF========
032+     float(2.3510579257887127E-38)
032-     float(-2.5961487387524236E+33)
========DONE========
FAIL Overflow in float to int cast [ext/exif/tests/float_cast_overflow.phpt]

I'm not super sure about the MySQL ones as they are maybe caused because I didn't think about how to set it up.

[guirish]

@Girgias I have already raised a PR #5380 to resolve MySQL specific TC failures.

[cmb69]

Regarding EXIF issues, see https://bugs.php.net/79336

[vibhutisawant]

Intl failures are due to the fact that the constuctor is failing e.g:

========DIFF========
001+ Fatal error: Uncaught IntlException: Constructor failed in /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php:4
001- bool(true)
002+ Stack trace:
003+ #0 /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php(4): ResourceBundle->__construct('es', '/home/travis/bu...')
004+ #1 {main}
005+   thrown in /home/travis/build/php/php-src/ext/intl/tests/resourcebundle_countable.php on line 4
========DONE========

@Girgias The above error is thrown as it fails to create a new Resource bundle Object. The resource file that the TC refers while calling the constructor is generated specifically for little endian arch.

Generate .res files for s390x using following commands:

icupkg -tb ext/intl/tests/_files/resourcebundle/root.res
icupkg -tb ext/intl/tests/_files/resourcebundle/es.res
icupkg -tb ext/intl/tests/_files/resourcebundle/res_index.res

Will it be possible to add the above changes in travis/compile.sh ? Such that it generates .res files for the respective arch.

[vibhutisawant]

The EXIF ones seems to related to overflow behaviour (and some cast with NaN which seem highly surprising)

========DIFF========
020+ float(8.914360180802324E-41)
020- float(NAN)
024+ float(8.914360180802324E-41)
024- float(NAN)
========DONE========
FAIL Bug #79046: NaN to int cast undefined behavior in exif [ext/exif/tests/bug79046.phpt]
========DIFF========
032+ float(2.3510579257887127E-38)
032- float(-2.5961487387524236E+33)
========DONE========
FAIL Overflow in float to int cast [ext/exif/tests/float_cast_overflow.phpt]

@Girgias wrt the above EXIF failures,

Following observations were made on Big endian System(s390x):
char is unsigned(0 to 255) on s390x, hence in function static float php_ifd_get_float(char *data) of
php-src/ext/exif/exif.c ,the 3rd byte (data[2]) is referring to -8(Hex:
fffffff8 ),where as its referred to 248 (Hex: f8) on s390x.
Could you please guide us on how can we proceed on this?

[Girgias]

The EXIF ones seems to related to overflow behaviour (and some cast with NaN which seem highly surprising)
========DIFF========
020+ float(8.914360180802324E-41)
020- float(NAN)
024+ float(8.914360180802324E-41)
024- float(NAN)
========DONE========
FAIL Bug #79046: NaN to int cast undefined behavior in exif [ext/exif/tests/bug79046.phpt]
========DIFF========
032+ float(2.3510579257887127E-38)
032- float(-2.5961487387524236E+33)
========DONE========
FAIL Overflow in float to int cast [ext/exif/tests/float_cast_overflow.phpt]

@Girgias wrt the above EXIF failures,

Following observations were made on Big endian System(s390x):
char is unsigned(0 to 255) on s390x, hence in function static float php_ifd_get_float(char *data) of
php-src/ext/exif/exif.c ,the 3rd byte (data[2]) is referring to -8(Hex:
fffffff8 ),where as its referred to 248 (Hex: f8) on s390x.
Could you please guide us on how can we proceed on this?

I don't really know EXIF, maybe @KalleZ has an idea?

[nikic]

I've fixed the exif issue and applied this change.

[nikic]

To reduce queue times for Travis, we might want to kill one of the x86_64 jobs, as we don't really need them.

[vibhutisawant]

@nikic Thanks for the fix.