Update FAQ: Shared does not cause memory barriers

[jpf91]

There's been a question about this in D.learn recently:
http://forum.dlang.org/post/apoqhsoqksuwieklcnjj@forum.dlang.org

If I remember correctly the final decision from 2012 was that shared variables do not automatically cause any memory barriers, but we never documented this:
http://forum.dlang.org/post/k7pn19$bre$1@digitalmars.com
http://forum.dlang.org/post/mailman.1904.1352922666.5162.digitalmars-d@puremagic.com

ping @WalterBright @andralex

[andralex]

I think we need to revisit this.

[JackStouffer]

@andralex I think shared needs to be revisited ;)

[d-random-contributor]

Hmm... was it initially meant so? Some people wanted it, but I don't think the compiler itself was meant to do it. Do you mean Bartosz proposal here?

[jpf91]

Hmm... was it initially meant so? Some people wanted it, but I don't think the compiler itself was meant to do it.

The FAQ entry changed by this pull request is currently:

Reading/writing shared data emits memory barriers to ensure sequential consistency (not implemented)

This is also mentioned in TDPL, chapter 13.12.1 Sequential consistency of shared data:

[...] To effect that, shared accesses must be surrounded by special machine code instructions called memory barriers, [...]

The discussion linked in the pull request description is more recent than TDPL though:
http://forum.dlang.org/post/k7pn19$bre$1@digitalmars.com

[andralex]

I think we made a mistake in the forum discussion. For now let's update this PR by making it state the current state of affairs: "Currently the compiler does not insert memory barriers around shared variables." Also please open a bugzilla issue to revisit this. Thanks.

[d-random-contributor]

Related issues are 14932 and 16198

[jpf91]

For now let's update this PR by making it state the current state of affairs:

Done.

Also please open a bugzilla issue to revisit this.

Do we need another issues or can we use 14932 and 16198 for this?

[JackStouffer]

I think 14932 suffices ;)

[andralex]

OK we're good with bugzilla. Thanks.