Disable date object related system calls by default.

[LaszloLango]

gettimeofday system call cause problems on multiple platforms, so better to disable by default.

Related issue: #923

JerryScript-DCO-1.0-Signed-off-by: László Langó llango.u-szeged@partner.samsung.com

[akosthekiss]

Issue analysis:

In jerry-core/ecma/builtin-objects/ecma-builtin-helpers-date.c, we rely on the second argument of gettimeofday(&tv, &tz) to determine the time zone. Unfortunately, Linux manuals state that "The use of the timezone structure is obsolete; the tz argument should normally be specified as NULL." Our bad luck is that qemu takes this seriously, and its gettimeofday syscall implementation simply avoids the second argument: it does not fill anything there, even if the pointer is non-NULL.

(Note: Real Linux kernels still provide valid tz info, that's why we haven't faced this problem until now. Moreover, our main target is a real device and not qemu emulation, of course. However, qemu is our only chance to get ARM tests work on Travis. So we should deal with the issue, IMHO.)

We could work around the trouble. At both places where we use the second argument, we use struct timezone tz;. If we would pre-initialize tz with zeros (= { 0, 0 };), then however we run jerry, we wouldn't find garbage in tz.tz_minuteswest and thus would be on the safe side.

Conclusions:

I wouldn't guard out date syscalls but would go with the initializations.

Notes:

• Changing ecma_raise_type_error ("gettimeofday failed") to ecma_number_make_nan () seems an unrelated change to me. Why's that?
• I'd definitely move struct timezone to sys/time.h since it's surely not the responsibility of ecma-builtin-helpers-date.c. (I haven't noticed its introduction in the last buildfix.)
• The above is kinda hipothetical, it hasn't beed tried out on jerry but on an example test app only.

[LaszloLango]

Changing ecma_raise_type_error ("gettimeofday failed") to ecma_number_make_nan () seems an unrelated change to me. Why's that?

The return value is ecma_number_t and ecma_raise_type_error returns ecma_value_t, so this is a bug and should be fixed.

[LaszloLango]

@akiss77, updated the PR. There are still build issues with external compilers. (Ex.: https://launchpad.net/gcc-arm-embedded/5.0/5-2015-q4-major/+download/gcc-arm-none-eabi-5_2-2015q4-20151219-linux.tar.bz2) So I still prefer to disable by default.

[akosthekiss]

Is there a way to enable the calls without editing config.h? If not, we wont be testing it "ever" -- neither with local precommit checks, nor on Travis. Which makes all gettimeofday-related code paths effectively dead. External tool chains are not the primary targets (strictly IMHO). If their users need tweaks, they shoud be the one to do edits to config.h.

[LaszloLango]

It looked like to me that is the common way. (see: CONFIG_VM_RUN_GC_AFTER_EACH_OPCODE) But I agree with you. A build option would be good.

[LaszloLango]

@akiss77, updated the patch. You can enable sys calls with DATE_SYS_CALLS=ON make... build option.

[akosthekiss]

The guards still make me sad, but LGTM (FWIW).

[zherczeg]

LGTM