Fixing remaining VC++ 2015 64-bit build breaks
VC++ 2015 64-bit builds trigger many new warnings about possibly
unintended truncation bugs. Some of the bugs have a very low signal-to-
noise ratio and this change suppresses them (4311 and 4312). Others seem
to find some real bugs so I am leaving them enabled for now (4302 and
4334), fixing the bugs, and adjusting the code to suppress the other
warnings.
The changes in shell_web_contents_view_delegate_win.cc,
ipc_perftest_support.cc, test_root_certs_win.cc, TracedLayoutObject.*
and TraceEvent.h appear to fix real truncation bugs.
Typical warnings are:
ipc\ipc_perftest_support.cc(351): warning C4334: '<<': result of 32-bit
shift implicitly converted to 64 bits (was 64-bit shift intended?)
net\cert\test_root_certs_win.cc(103): warning C4302: 'reinterpret_cast':
truncation from 'LPCSTR' to 'uint32_t'
The main pattern is that VC++ 2015 is suspicious of code that converts a
64-bit pointer to a 32-bit integer. If the code converts from a pointer
to a 64-bit integer, and then to a 32-bit integer, separating the type
change from the truncation, then it assumes that the truncation is
intentional. This seems like a reasonable heuristic. The warnings in the
generated code were suppressed because they aren't bugs and because
fixing gperf.exe (untouched since 2004) does not seem worthwhile.
The warnings from gperf generated code look like:
out\release_x64\gen\blink\core\csspropertynames.cpp(2914): warning C4302:
'type cast': truncation from 'char (*)[28]' to 'long'
out\release_x64\gen\blink\core\cssvaluekeywords.cpp(3409): warning C4302:
'type cast': truncation from 'char (*)[4]' to 'long'
Note that Windows HANDLE types are a wrapper around a pointer, but due
to WOW64 (32-bit process on 64-bit Windows) compatibility they are
always 32-bit values that can safely be truncated, then sign-extended
back. Unfortunately Microsoft does not supply a safe-truncation macro.
https://msdn.microsoft.com/en-us/library/windows/desktop/aa384203%28v=vs.85%29.aspx
Also a few uncommented warning-disables in common.gypi were commented.
Setting NOTRY=true because the change is hitting timeouts. See
crbug.com/567377 and comment #119 for details.
NOTRY=true
BUG=440500
Review URL: https://codereview.chromium.org/1422773008
Cr-Commit-Position: refs/heads/master@{#364264}
26 files changed
