• Jacob Carlborg (12/12) Mar 21 2020 So far I've been building LDC against upstream LLVM 9.0.0. When I build
• kinke (4/6) Mar 21 2020 There's only one LDC-specific Mach-O related change potentially
• Jacob Carlborg (5/10) Mar 22 2020 Yeah, that's it. If I run with `--preserve-dwarf-line-section=false` the...
• kinke (8/11) Mar 22 2020 Well, it was clear from the beginning that that's a hack, and the

Jacob Carlborg <doob me.com> writes:

So far I've been building LDC against upstream LLVM 9.0.0. When I build 
with the LDC fork of LLVM, 9.0.1 I a bunch of the following warnings 
when compiling the test runners for iOS:

ld: warning: pointer not aligned at address 0x10321E776 (anon + 1270 
from 
~/development/d/ios-druntime/ldc-build-runtime.tmp/lib/libphobos2-ldc-unittest-debug.a(uni.o))

ld: warning: pointer not aligned at address 0x1033AAFB3 (anon + 110 from 
lib/libdruntime-ldc-debug.a(ti_Areal.o))

I'm not sure if it's because it's LLVM 9.0.1 or if it's because the LDC 
fork.

-- 
/Jacob Carlborg

kinke <noone nowhere.com> writes:

On Saturday, 21 March 2020 at 19:20:22 UTC, Jacob Carlborg wrote:
 I'm not sure if it's because it's LLVM 9.0.1 or if it's because 
 the LDC fork.

There's only one LDC-specific Mach-O related change potentially 
related to this, 
https://github.com/ldc-developers/llvm-project/commit/110deda1bc1cf195983fe
8c1107886057987955. [The effect of which can be tested by hacking LDC, grep for
`emitMachODwarfLineAsRegularSection`.]

Jacob Carlborg <doob me.com> writes:

On 2020-03-21 20:29, kinke wrote:

 There's only one LDC-specific Mach-O related change potentially related 
 to this, 
 https://github.com/ldc-developers/llvm-project/commit/110deda1bc1cf195983fe
8c1107886057987955. 
 [The effect of which can be tested by hacking LDC, grep for 
 `emitMachODwarfLineAsRegularSection`.]

Yeah, that's it. If I run with `--preserve-dwarf-line-section=false` the 
problem is gone. That's unfortunate :(. It seem to run properly anyway.

-- 
/Jacob Carlborg

kinke <noone nowhere.com> writes:

On Sunday, 22 March 2020 at 08:01:33 UTC, Jacob Carlborg wrote:
 Yeah, that's it. If I run with 
 `--preserve-dwarf-line-section=false` the problem is gone. 
 That's unfortunate :(. It seem to run properly anyway.

Well, it was clear from the beginning that that's a hack, and the 
proper fix needs to be in druntime, not relying on extra 
preserved sections for backtrace lineinfos 
(https://issues.dlang.org/show_bug.cgi?id=20510).

On the plus side, it should only affect builds with debuginfos 
and can be disabled without hacking LDC (thx, I forgot about that 
switch).