[llvm-bugs] [Bug 27887] New: Arguments marked as ns_consumed should not be released if forwarded to another ns_consumed argument

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=27887

            Bug ID: 27887
           Summary: Arguments marked as ns_consumed should not be released
                    if forwarded to another ns_consumed argument
           Product: clang
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: sdefresne at google.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Trying to use "ns_consumed" in a template class, I found that a function taking
an argument marked as "ns_consumed" and forwarding it to another function
taking it as "ns_consumed" too behave differently in constructor and plain
functions.

With free functions, no memory management happens (as expected):

$ cat x.mm
void foo(id __attribute__((ns_consumed)) x);
void bar(id __attribute__((ns_consumed)) x) {
  foo(x);
}
$ llvm-build/Release+Asserts/bin/clang++ -S -o - x.mm -fomit-frame-pointer
-fobjc-arc -Os -fno-exceptions --std=c++11
    .section    __TEXT,__text,regular,pure_instructions
    .macosx_version_min 10, 11
    .globl    __Z3barU11ns_consumedP11objc_object
__Z3barU11ns_consumedP11objc_object:    ## @_Z3barU11ns_consumedP11objc_object
    .cfi_startproc
## BB#0:                                ## %entry
    pushq    %rax
Ltmp0:
    .cfi_def_cfa_offset 16
    callq    __Z3fooU11ns_consumedP11objc_object
    popq    %rax
    retq
    .cfi_endproc

    .section    __DATA,__objc_imageinfo,regular,no_dead_strip
L_OBJC_IMAGE_INFO:
    .long    0
    .long    64


.subsections_via_symbols

However, for constructors (I've not checked for other methods), there is a call
to objc_release that I do not expect (as there are no corresponding
obj_retain):

$ cat y.mm 
struct a {
  a(id __attribute__((ns_consumed)) x) noexcept;
};
struct b : a {
  b(id __attribute__((ns_consumed)) x);
};
b::b(id __attribute__((ns_consumed)) x) : a(x) {}
$ llvm-build/Release+Asserts/bin/clang++ -S -o - y.mm -fomit-frame-pointer
-fobjc-arc -Os -fno-exceptions --std=c++11
    .section    __TEXT,__text,regular,pure_instructions
    .macosx_version_min 10, 11
    .globl    __ZN1bC2EU11ns_consumedP11objc_object
    .p2align    1, 0x90
__ZN1bC2EU11ns_consumedP11objc_object:  ##
@_ZN1bC2EU11ns_consumedP11objc_object
    .cfi_startproc
## BB#0:                                ## %entry
    jmp    __ZN1aC2EU11ns_consumedP11objc_object ## TAILCALL
    .cfi_endproc

    .globl    __ZN1bC1EU11ns_consumedP11objc_object
    .p2align    1, 0x90
__ZN1bC1EU11ns_consumedP11objc_object:  ##
@_ZN1bC1EU11ns_consumedP11objc_object
    .cfi_startproc
## BB#0:                                ## %entry
    pushq    %rbx
Ltmp0:
    .cfi_def_cfa_offset 16
Ltmp1:
    .cfi_offset %rbx, -16
    movq    %rsi, %rbx
    callq    __ZN1aC2EU11ns_consumedP11objc_object
    movq    %rbx, %rdi
    popq    %rbx
    jmpq    *_objc_release at GOTPCREL(%rip) ## TAILCALL
    .cfi_endproc

    .section    __DATA,__objc_imageinfo,regular,no_dead_strip
L_OBJC_IMAGE_INFO:
    .long    0
    .long    64


.subsections_via_symbols

I would expect the code generated to be the same here as is generated with
-fno-objc-arc.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160526/916ad263/attachment.html>