Win64 failing unittests

[Trass3r]

Let's collect information about the failing unittests.

• _d_arraycatnT contains varargs and crashes. Responsible for rt.lifetime, core.time, std.array, std.conv,... probably most of them.
  Potentially related to Vararg functions does not compile on x86_64 #702 core.stdc.stdarg doesn't work on 64 bits #73 Crash when compiling dbox (also compiling forever on release) #699
• std.complex test fails @ line 480 with c1^^c2 cause of some weirdness in std.math atan2(real y, real x) leading to arg(1+i) to be 0. [Win64] real constants parsed as 0 #761
• core.thread is weird crash in ntdll.dll!RtlDispatchException()

[Trass3r]

http://msdn.microsoft.com/en-us/library/dd2wa36c.aspx

[kinke]

Good idea. Current state for those interested:

57 failures:

core.thread (SEGFAULT)
core.time (SEGFAULT)
rt.lifetime (SEGFAULT)
std.algorithm (SEGFAULT)
std.array (SEGFAULT)
std.complex (Failed)
std.concurrency (SEGFAULT)
std.conv (SEGFAULT)
std.csv (Failed)
std.datetime (SEGFAULT)
std.getopt (SEGFAULT)
std.math (Failed)
std.numeric (Failed)
std.outbuffer (Failed)
std.parallelism (SEGFAULT)
std.path (SEGFAULT)
std.process (SEGFAULT)
std.regex (SEGFAULT)
std.socket (Failed)
std.stdio (Failed)
std.string (SEGFAULT)
std.uni (SEGFAULT)
std.uri (SEGFAULT)
std.variant (Failed)
std.xml (SEGFAULT)
std.zlib (SEGFAULT)
std.net.isemail (SEGFAULT)
std.internal.math.errorfunction (Failed)
std.internal.math.gammafunction (Failed)
core.time-debug (SEGFAULT)
rt.lifetime-debug (SEGFAULT)
std.algorithm-debug (SEGFAULT)
std.array-debug (SEGFAULT)
std.complex-debug (Failed)
std.concurrency-debug (SEGFAULT)
std.conv-debug (SEGFAULT)
std.csv-debug (Failed)
std.datetime-debug (SEGFAULT)
std.getopt-debug (SEGFAULT)
std.json-debug (Failed)
std.math-debug (Failed)
std.numeric-debug (Failed)
std.outbuffer-debug (Failed)
std.parallelism-debug (SEGFAULT)
std.path-debug (Failed)
std.process-debug (SEGFAULT)
std.regex-debug (SEGFAULT)
std.socket-debug (Failed)
std.stdio-debug (Failed)
std.uni-debug (SEGFAULT)
std.uri-debug (SEGFAULT)
std.variant-debug (SEGFAULT)
std.xml-debug (SEGFAULT)
std.zlib-debug (SEGFAULT)
std.net.isemail-debug (SEGFAULT)
std.internal.math.errorfunction-debug (Failed)
std.internal.math.gammafunction-debug (Failed)

Kai's latest EH fixes have reduced the number further.
That vararg bug looks very promising...

[Trass3r]

Added #761.

[kinke]

The number is now down to 41: ldc-developers/druntime#14 (comment)

[kinke]

Update:
39 failures with LLVM 3.7 (d89ac8f158976eaa9f9bcfba88f8a679771704f8), LDC head (with /LARGEADDRESSAWARE:NO adaptations) and VS 2013/Win8 x64:

core.thread
std.algorithm
std.complex
std.conv
std.csv
std.datetime
std.math
std.numeric
std.parallelism
std.path
std.process
std.regex
std.stream
std.uni
std.uri
std.zlib
std.net.isemail
std.internal.math.errorfunction
std.internal.math.gammafunction
std.algorithm-debug
std.complex-debug
std.conv-debug
std.csv-debug
std.datetime-debug
std.json-debug
std.math-debug
std.numeric-debug
std.parallelism-debug
std.path-debug
std.process-debug
std.regex-debug
std.socket-debug
std.stream-debug
std.uni-debug
std.uri-debug
std.zlib-debug
std.net.isemail-debug
std.internal.math.errorfunction-debug
std.internal.math.gammafunction-debug

[kinke]

I'll go ahead with VS 2015 CTP 6 ;) - not too bad, 58 failures:

         19 - core.thread (SEGFAULT)
        147 - std.algorithm (Failed)
        148 - std.array (Failed)
        151 - std.bigint (Failed)
        156 - std.conv (Failed)
        157 - std.cstream (Failed)
        158 - std.csv (Failed)
        159 - std.datetime (Failed)
        161 - std.encoding (Failed)
        164 - std.format (Failed)
        168 - std.math (Failed)
        174 - std.parallelism (SEGFAULT)
        175 - std.path (Failed)
        176 - std.process (Failed)
        179 - std.regex (Failed)
        181 - std.socket (Failed)
        184 - std.stdio (Failed)
        186 - std.stream (SEGFAULT)
        187 - std.string (Failed)
        190 - std.traits (Failed)
        194 - std.uni (Failed)
        195 - std.uri (Failed)
        201 - std.zlib (Failed)
        209 - std.digest.crc (Failed)
        211 - std.digest.md (Failed)
        212 - std.digest.ripemd (Failed)
        213 - std.digest.sha (Failed)
        215 - std.net.isemail (SEGFAULT)
        221 - std.internal.math.gammafunction (Failed)
        390 - std.algorithm-debug (Failed)
        391 - std.array-debug (Failed)
        394 - std.bigint-debug (Failed)
        399 - std.conv-debug (Failed)
        400 - std.cstream-debug (Failed)
        401 - std.csv-debug (Failed)
        402 - std.datetime-debug (Failed)
        404 - std.encoding-debug (Failed)
        407 - std.format-debug (Failed)
        410 - std.json-debug (Failed)
        411 - std.math-debug (Failed)
        417 - std.parallelism-debug (SEGFAULT)
        418 - std.path-debug (Failed)
        419 - std.process-debug (Failed)
        422 - std.regex-debug (Failed)
        424 - std.socket-debug (Failed)
        427 - std.stdio-debug (Failed)
        429 - std.stream-debug (SEGFAULT)
        430 - std.string-debug (Failed)
        433 - std.traits-debug (Failed)
        437 - std.uni-debug (Failed)
        438 - std.uri-debug (Failed)
        444 - std.zlib-debug (Failed)
        452 - std.digest.crc-debug (Failed)
        454 - std.digest.md-debug (Failed)
        455 - std.digest.ripemd-debug (Failed)
        456 - std.digest.sha-debug (Failed)
        458 - std.net.isemail-debug (SEGFAULT)
        464 - std.internal.math.gammafunction-debug (Failed)

with ldc-developers/phobos#6 and ldc-developers/druntime#17 (don't forget to uncomment version = LDC_MSVCRT14; in the druntime files).

A new common issue seems to have arised (assertion in CRT) which probably accounts for most differences compared to MSVCRT 12.
MSVCRT 14 fixes highly problematic #761. We can't fix that easily otherwise.

/edit: I think I've fixed that issue, feof() didn't work as stdio.h has changed more than anticipated. I've silently updated the druntime PR and am currently rebuilding the tests.
/edit2: Nope that didn't fix the main issue.

[kinke]

Now down to 39 failures with #856 and MSVCRT 14 by simply disabling a printf() call in test_runner.d after a successful unit-test run, the printf() call inside the try-block. Apparently there's some serious corruption going on as otherwise I get a MSVCRT assertion failure

(fh >= 0 && (unsigned)fh < (unsigned)_nhandle)

in isatty() called by printf() which probably checks that stdout points to a valid handle.
By disabling the try-catch-block altogether and instead simply invoking fp(), a bunch of randomly selected failing Phobos-debug tests according to VS debugger all fail in ldc.eh2._d_eh_personality:364 due to an access violation:

if (_d_isbaseof(excobj.classinfo, catch_ci))

with excobj being null. The call stack is full of _d_eh_personality() frames.
I guess exception handling is still the #1 blocker for Win64.

Failing tests:

core.thread (SEGFAULT)
std.algorithm (SEGFAULT)
std.csv (Failed)
std.datetime (SEGFAULT)
std.math (Failed)
std.parallelism (SEGFAULT)
std.path (SEGFAULT)
std.process (SEGFAULT)
std.regex (SEGFAULT)
std.stream (SEGFAULT)
std.uni (SEGFAULT)
std.uri (SEGFAULT)
std.zlib (SEGFAULT)
std.net.isemail (SEGFAULT)
std.internal.math.gammafunction (Failed)
std.algorithm-debug (SEGFAULT)
std.array-debug (SEGFAULT)
std.bigint-debug (SEGFAULT)
std.csv-debug (Failed)
std.encoding-debug (SEGFAULT)
std.json-debug (Failed)
std.math-debug (Failed)
std.parallelism-debug (SEGFAULT)
std.path-debug (SEGFAULT)
std.process-debug (SEGFAULT)
std.regex-debug (SEGFAULT)
std.socket-debug (Failed)
std.stream-debug (SEGFAULT)
std.string-debug (SEGFAULT)
std.traits-debug (SEGFAULT)
std.uni-debug (SEGFAULT)
std.uri-debug (SEGFAULT)
std.zlib-debug (SEGFAULT)
std.digest.crc-debug (SEGFAULT)
std.digest.md-debug (SEGFAULT)
std.digest.ripemd-debug (SEGFAULT)
std.digest.sha-debug (SEGFAULT)
std.net.isemail-debug (SEGFAULT)
std.internal.math.gammafunction-debug (Failed)

/edit: If I abort() right before _d_eh_personality:364 if excobj is null, all of above's SEGFAULTs except for core.thread are converted to aborts. I.e., this is crucial. printf() is not usable at that point, causing the same CRT assertion.

[kinke]

Only remaining failing unittest nowadays is core.thread, which fails in release in line https://github.com/ldc-developers/druntime/blob/ldc/src/core/thread.d#L5467, a Win64-specific test.

[kinke]

The status is still the same. #1103 points to the single remaining 'issue' incl. details, so I'm going to close this.