[lld] [lld][macho] Support order cstrings with -order_file_cstring (PR #140307)

[Snehasish Kumar via llvm-commits]

================
@@ -0,0 +1,226 @@
+# RUN: rm -rf %t; split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin  %t/test.s -o %t/test.o
+# RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %t/more-cstrings.s -o %t/more-cstrings.o
+
+# RUN: %lld --deduplicate-strings -arch arm64 -lSystem -e _main -o %t/test-0 %t/test.o %t/more-cstrings.o
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/test-0 | FileCheck %s --check-prefix=ORIGIN_SYM
+# RUN: llvm-objdump --macho --section="__TEXT,__cstring" %t/test-0 | FileCheck %s --check-prefix=ORIGIN_SEC
+
+# RUN: %lld --deduplicate-strings -arch arm64 -lSystem -e _main -o %t/test-1 %t/test.o %t/more-cstrings.o -order_file %t/ord-1
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/test-1 | FileCheck %s --check-prefix=ONE_SYM
+# RUN: llvm-objdump --macho --section="__TEXT,__cstring" %t/test-1 | FileCheck %s --check-prefix=ONE_SEC
+
+
+# RUN: %lld --deduplicate-strings -arch arm64 -lSystem -e _main -o %t/test-2 %t/test.o %t/more-cstrings.o -order_file %t/ord-2
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/test-2 | FileCheck %s --check-prefix=TWO_SYM
+# RUN: llvm-objdump --macho --section="__TEXT,__cstring" %t/test-2 | FileCheck %s --check-prefix=TWO_SEC
+
+# RUN: %lld --deduplicate-strings -arch arm64 -lSystem -e _main -o %t/test-3 %t/test.o %t/more-cstrings.o -order_file %t/ord-3
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/test-3 | FileCheck %s --check-prefix=THREE_SYM
+# RUN: llvm-objdump --macho --section="__TEXT,__cstring" %t/test-3 | FileCheck %s --check-prefix=THREE_SEC
+
+# RUN: %lld --deduplicate-strings -arch arm64 -lSystem -e _main -o %t/test-4 %t/test.o %t/more-cstrings.o -order_file %t/ord-4
+# RUN: llvm-nm --numeric-sort --format=just-symbols %t/test-4 | FileCheck %s --check-prefix=FOUR_SYM
+# RUN: llvm-objdump --macho --section="__TEXT,__cstring" %t/test-4 | FileCheck %s --check-prefix=FOUR_SEC
+# RUN: llvm-readobj --string-dump=__cstring %t/test-4 | FileCheck %s --check-prefix=FOUR_SEC_ESCAPE
+
+
+# We expect:
+# 1) Covered cstring symbols are reordered
+# 2) the rest of the cstring symbols remain original relative order within the cstring section
+
+# ORIGIN_SYM: _local_foo1
+# ORIGIN_SYM: _globl_foo2
+# ORIGIN_SYM: _local_foo2
+# ORIGIN_SYM: _bar
+# ORIGIN_SYM: _baz
+# ORIGIN_SYM: _baz_dup
+# ORIGIN_SYM: _bar2
+# ORIGIN_SYM: _globl_foo3
+
+# ORIGIN_SEC: foo1
+# ORIGIN_SEC: foo2
+# ORIGIN_SEC: bar
+# ORIGIN_SEC: baz
+# ORIGIN_SEC: bar2
+# ORIGIN_SEC: foo3
+
+# original order, but only parital covered
----------------
snehasish wrote:

I asked about the RFC since @ellishg mentioned it earlier (so it would be about this patch).

> I think it makes sense to put `cstr:` after the cpu type and file separately: `[<CPU>:][<OBJ>.o:][cstr:]<symbol>`
> 
> Maybe we should put up an RFC on https://discourse.llvm.org/

^ This is the comment I am referring to.

In my opinion the changes and limited and have precedent in terms of the prefix based format so it isn't necessary.

> (Sorry was having a couple of sick days, probably going to submit the draft patch for the binary access profiling early next week)

Thanks! No rush, please tag @teresajohnson @mingmingl-llvm and me when you are ready.

https://github.com/llvm/llvm-project/pull/140307