fix Issue 16408 - [REG2.065] left-to-right order of evaluation of fun… #6705
Conversation
|
WalterBright
force-pushed
the
fix16408
branch
13 times, most recently
from
2965cfd to
e6a4812
Compare
…ction arguments not consistently followed
|
This PR makes left-to-right evaluation of function arguments much more consistent than before. While this is in more conformance with the D spec, it may break some existing code that inadvertently had a dependency of r-t-l evaluation. It is impractical for the compiler to detect this (good 'ole halting problem). |
|
Can we then close the order of evaluation matter, or there are a few stragglers left? |
| @@ -182,10 +182,10 @@ elem *callfunc(Loc loc, | |||
| ty = ec.Ety; | |||
| if (fd) | |||
| ty = toSymbol(fd).Stype.Tty; | |||
| reverse = tyrevfunc(ty); | |||
| reverse = tyrevfunc(ty); // left-to-right parameter evaluation (TYnpfunc, TYjfunc, TYfpfunc, TYf16func) | |||
| elem*[5] elems_array = void; | ||
| import core.stdc.stdlib : malloc, free; | ||
| auto elems = (n <= 5) ? elems_array.ptr : cast(elem**)malloc(arguments.dim * (elem*).sizeof); | ||
| assert(elems); |
There was a problem hiding this comment.
Would be nicer to use this instead of relying on code downstream:
scope(exit) if (n > 5) free(elems);There was a problem hiding this comment.
I've avoided that form because the intervening code is not nothrow, meaning it sets up an EH frame even though the intervening code will never throw.
| * D presumes left-to-right argument evaluation, but we're evaluating things | ||
| * right-to-left here. | ||
| * 1. determine if this matters | ||
| * 2. fix it if it does |
There was a problem hiding this comment.
It does matter... best to keep to the spec and keep the spec simple
There was a problem hiding this comment.
I meant that foo(3,4) does not have order of evaluation issues.
|
There may be a couple other OOE issues, but this should fix the biggie. |
|
Auto-merge toggled on |
|
This apparently introduced a regression Issue 17356 – [Reg 2.075] __simd_sto no longer executed @WalterBright, yet another backend issue sabotaging the arrayops rewrite :/. |
|
@WalterBright r-t-l order makes more sense for function calls. |
…ction arguments not consistently followed