The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9c995a82acfcfc47c8650c5edea8e5e6c28d7893

commit 9c995a82acfcfc47c8650c5edea8e5e6c28d7893
Author: James Darpinian <jdarpinian@chromium.org>
Date: Tue Dec 11 22:00:24 2018

gpu: Change CHECK to DCHECK in transfer buffer

As requested in the review of https://crrev.com/c/1336753. After
https://crbug.com/913996 is fixed we can remove
outstanding_result_pointer_ entirely when DCHECKs are disabled.

It also turns out that we can't use the string argument to ASSERT_DEATH
to detect a CHECK message in a test because the message strings are
stripped in official builds.

Bug: 905890,  913421 , 913996
Change-Id: Ia7df19057564ef7a4bbbab9ba36f583c6e6bca0f
Reviewed-on: https://chromium-review.googlesource.com/c/1370484
Commit-Queue: James Darpinian <jdarpinian@chromium.org>
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#615680}
[modify] https://crrev.com/9c995a82acfcfc47c8650c5edea8e5e6c28d7893/gpu/command_buffer/client/transfer_buffer.cc
[modify] https://crrev.com/9c995a82acfcfc47c8650c5edea8e5e6c28d7893/gpu/command_buffer/client/transfer_buffer_unittest.cc

If you get the same error in android-binary-size the common factor is probably not jumbo but some "race condition" that gets triggered when you shake the build up a bit.

The way we define DCHECK, I actually don't know how this should ever work.

base/logging.h says:

// DCHECK et al. make sure to reference |condition| regardless of
// whether DCHECKs are enabled; this is so that we don't get unused
// variable warnings if the only use of a variable is in a DCHECK.
// This behavior is different from DLOG_IF et al.

and defines DCHECK, when turned off to:

#define DCHECK(condition) EAT_STREAM_PARAMETERS << !(condition)

EAT_STREAM_PARAMETERS is 
#define EAT_STREAM_PARAMETERS \
  true ? (void)0              \
       : ::logging::LogMessageVoidify() & (*::logging::g_swallow_stream)

Which I clearly don't understand, but either way, the variable will remain in the code that the compiler sees.

So that 
#if DCHECK_IS_ON()
int foo = 1;
#endif
DCHECK(foo);

becomes 

true ? (void)0              \
       : ::logging::LogMessageVoidify() & (*::logging::g_swallow_stream) << !(foo);

where foo is an undeclared symbol.  So how does this not cause compilation errors in most builds? And what is different with linux-jumbo-rel and android-binary-size?