place statements array inside CompoundStatements, it is non-empty most of the time.

[rainers]

reduces size and avoids an extra allocation.

Also avoids the indirection on usage and makes the code a bit cleaner.

The downside is that constructing the CompoundStatement with an array of statements has to move the contents of the array if we don't want the extra copy - which might not be obvious from the code. This could be done a bit nicer with rvalues, but the bootstrap compilers don't have that.

[github-actions]

DMD perf check

Metric	Base	PR	delta
compile hello.d (instr)	215.8 M	215.7 M	-0.01%
compile hello.d -O (instr)	234.3 M	234.3 M	0.00%
compile Phobos (instr)	5,080.4 M	5,077.5 M	-0.06%
dmd binary size (stripped)	6.88 MB	6.91 MB	+0.34%
hello binary size	0.72 MB	0.72 MB	0.00%
peak RSS (compile hello.d)	44 MB	44 MB	+0.02%
peak RSS (compile Phobos)	633 MB	631 MB	-0.31%

[rainers]

The downside is that constructing the CompoundStatement with an array of statements has to move the contents of the array if we don't want the extra copy - which might not be obvious from the code. This could be done a bit nicer with rvalues, but the bootstrap compilers don't have that.

I could workaround dmd 2.079 not doing the return-value-optimizations by an additional ctor, so Statements are now explicitly moved into CompoundStatements and UnrolledLoopStatements.

[thewilsonator]

Is this good to go?

[rainers]

Is this good to go?

I wasn't too happy about statementSemanticVisit.visitScope moving the statements out of a CompoundStatement after a call to flatten instead of just sharing them with the original statement, just to notice that the new Statement is unnecessary in that case anyway, see https://github.com/dlang/dmd/pull/23421/changes#diff-11fb0212c55c05ed6e743e6776abb3c56996381760c8725255c622f96f5d9d4bR561
So a couple of more MBs saved.

If the changes are not considered too disturbing, I'm fine with it now.

[thewilsonator]

If you can version out a fix that is not too intrusive for 2.079 that would be great.

[rainers]

If you can version out a fix that is not too intrusive for 2.079 that would be great.

Ok, done. The single Array!T.move() function is not a lot shorter, though.