Bug 12308 – Request pragma for very thin struct wrappers

Status
NEW
Severity
enhancement
Priority
P4
Component
dmd
Product
D
Version
D2
Platform
All
OS
All
Creation time
2014-03-06T20:04:13Z
Last change time
2024-12-13T18:17:57Z
Assigned to
No Owner
Creator
Adam D. Ruppe
Moved to GitHub: dmd#17645 →

Comments

Comment #0 by destructionator — 2014-03-06T20:04:13Z
In short, I want codegen for pragma(thin_struct) struct S { int; } to be identical in every way as for int; Especially, struct member functions just get the value in the register instead of a pointer, and returning structs is done in the register instead of via hidden pointer. http://forum.dlang.org/thread/[email protected]?page=2#post-jklqyrulonhxwrucvbil:40forum.dlang.org
Comment #1 by bearophile_hugs — 2014-03-07T02:21:06Z
Walter seems clearly against this idea: --------------------- http://forum.dlang.org/post/[email protected] Walter Bright: Posted in reply to Adam D. Ruppe On 3/6/2014 8:01 PM, Adam D. Ruppe wrote: > BTW you know what would help this? A pragma we can attach to a struct which > makes it a very thin value type. I'd rather fix the compiler's codegen than add a pragma. --------------------- http://forum.dlang.org/post/[email protected] Walter Bright: On 3/6/2014 10:12 PM, H. S. Teoh wrote: > From what I understand, structs are *supposed* to be thin value types. I > would say that if a struct is under a certain size (determined by the > compiler), and doesn't have complicated semantics like dtors and stuff > like that, then it should be treated like a POD (passed in registers, > etc). Yes, that's right. ---------------------
Comment #2 by destructionator — 2014-03-07T07:18:14Z
I don't think it has been explained fully yet. A struct with just an int is not fully interchangeable for an int because they are returned differently by functions, including in C. So this isn't really a matter of fixing anything, since it isn't broken, it is a different spec. indeed, since it changes the calling convention, how would this mangle? I'd say the same as the type it wraps. The struct just disappears as far as the backend is concerned. A few other points in the thread: couldn't the hidden this be changed to pass by value? Maybe, though taking the address of this wouldn't work, I think. A* ptr() { return &this; } works now, just returning EAX (which is the hidden this pointer in the first place). Trying that with a regular int is (awesomely!) an error: int* foo(int a) { return &a; } test56.d(17): Error: escaping reference to local a * * * Could we just change the D calling convention to pass small structs by value* while keeping extern(C) the same for compatibility? Yes, I think we can, and I think it would be a win. But I'd still like it to be available in extern(C) too for cases like library typedef, where we want it to look different on the D side, but remain ABI identical to the naked type. * Note: this is already what happens for most function arguments: void test(A a); calling that looks like: mov EAX, [a]; call test; Which is identical to test(int a). The difference is returning a value and the hidden this pointer. Returning a value is done by hidden pointer too.
Comment #3 by robert.schadek — 2024-12-13T18:17:57Z
THIS ISSUE HAS BEEN MOVED TO GITHUB https://github.com/dlang/dmd/issues/17645 DO NOT COMMENT HERE ANYMORE, NOBODY WILL SEE IT, THIS ISSUE HAS BEEN MOVED TO GITHUB