[OT] Dear Google Cloud: Your Deprecation Policy is Killing You
user1234 at 12.de
Thu Aug 20 11:31:04 UTC 2020
On Thursday, 20 August 2020 at 06:30:11 UTC, Mathias LANG wrote:
> On Wednesday, 19 August 2020 at 15:59:52 UTC, H. S. Teoh wrote:
>> "Went over very smoothly" is a total mockery of the reality of
>> the situation.
> I concur. And to follow the "Bugzilla or GTFO" rule:
From what I've read recently, the problem is not the dprecation.
The problem is that a deprecation has been used to hide a
limitation of the backend.
The problem is that DMD backend generates instructions for 32bit
R/M or 64 bits only R/M only. It does not care if the source type
is 8bit or 16 bit. The source type is supposed to be at least a
32 bits int. so the big problem is... if you try to write an
unalagined data, the instruction will be wrong. If you write an
ushort that starts two bytes before a new system aligned address
then the two first bytes of the next system aligned address will
be eventually corrupted. The 16 bits versions of the instructions
are never generated, that is the real problem. D backend wants
everything to be 8 or 4 aligned.
More information about the Digitalmars-d