Sorry for the tardiness of this...

1. Could the the type change from std::string be to the new State type introduced (rather than int)
2. Have all State variables default initialized to a value that would cause the test to fail.

Crap your right...I might revisit this (as I think that's a bad move on my part).
I'd still like to give a moment for anyone else to chime in; otherwise, it shouldn't be a reason to hold this up.

@lhames, any thoughts?

I take it your using the MSVC generator?
I use ninja (which is a single-config build) all the time and it puts the the files into unittests/Support/DynamicLibrary ?
I'll try to spin up a Windows build in the next couple of days, but I'm not foreseeing my opinion being changed.

I'm not a fan of writing test stuff into <buildroot>/lib.

Thanks.

Just maybe add a comment similar to the last line of the summary

Sorry 'bout that...
(CMake and quoting make dull boys)

Shouldn't the Translation Unit be responsible for knowing what it needs rather than hoping it's already been used somewhere up the chain.
Perhaps std::unique_ptr and std::make_pair are too low level and one can assume, but D44509 seems like a possibly overzealous application?

In D30709#1019309, @LepelTsmok wrote:

I think what you're trying to do just isn't unsupported at this time.

So it is supported?

I think what you're trying to do just isn't unsupported at this time.
Why not just use the JIT to compile the source for the on disk object at runtime?

It's a little hard to understand the issue. To be clear:
This patch is fixing an instance that used to crash?
However there is a bit of manual setup for the JIT you are still not happy with?

Yes, thanks.
Just waiting till I had a moment to watch the new test on the bots.

Changed name of function to write32BitOffset.

Ping

Ping.

Ping.

Added test case for IMAGE_REL_AMD64_ADDR32NB with JIT and requested formatting changes.

I can commit this, but it should probably be merged into release_50 as well (?)

LGTM.

I'm wondering if the #ifdef is now a dead code path entirely.
Do you have any insight into how ARM is getting __atomic_load_n and x86/64 not?

Done, thanks. Would it be possible to look at D30709 as this depends on that. (ELF does not like Windows exceptions).

Use include guards, not pragma once and remove commented CMake debug lines.

Ping,

Ping,

Ping.

Ping.

Ping.

Done, and changed to 'windows.h', but blocked mingw from even attempting SEH for now.

Blocked Win64 SEH for MingW entirely.
Theoretically it should be possible to throw a SEH from within mingw, but but I'm thinking clang may be generating exception handlers for the gcc runtime in that case.
I'll leave it up to someone with interest in such a mixture to figure that tout.

Added support for IEC, SI, and Customary output.

@joerg Any further thoughts/desire to review this?
Finishing this off would be nice and at least opens the door to using RTLD_LOCAL.

@compnerd is it possible to restart this.
I added an example to clang that verifies this works in D35103.
Just don't know how to trigger the relocation types in LLVM.

@ruiu
What about an optional argument to specific IEC suffixes.
Using IEC by default is fine by me, just that there is a way to not do so.

Ping.
@lhames Is there any chance to get this in before 5.0...it removes the need to store this same exact information 2x in our code base.

I appreciate the differences, but this is specifically to output an amount of memory in a human readable form.
Using KiB, MiB, etc may be pedantically correct, but it limits the usefulness of this to format text for a user who may have little knowledge of the differences between SI and IEC suffixes.
Every operating system I'm aware of does it this way and having output correlate to how the OS reports memory is more important.

Fix typo in documentation.

Updated

Ok,

Use to std::pair instead of PointerIntPair, it's lighter for the temporary.

@v.g.vassilev @pcanal :
Can you provide a concrete example of a test in ROOT that this patch does not solve?

Ping.

This has nothing to do with fixing your issue and is solely to allow clients to load via RTLD_LOCAL.
I don't see how D33529 could have broken anything as it is not committed yet. D33529 attempts to address what I understand is your issue.
If it's not working perhaps you could explain or provide a bit more info what doesn't work...(on that thread not here).
Perhaps @pcanal can help with that?

Typo & reword documentation.

From what I can tell, in LLVM the clients of this are solely clang and lldb.
Nothing should break for them as they explicitly search single libraries for symbols.

ping

Works fine here against master, though D33658 did have a conflict.

Include PointerIntPair.h for modules build & resolved conflict.

Ping.

Ping.

Ping.

Ping.

From the summary:

Fixes an issue using RegisterStandardPasses from a statically linked object before PassManagerBuilder::addGlobalExtension is called from a dynamic library.

Format.

I'm missing the *why* using a "real function" (I guess you meant "function pointer") matter?

Remove spurious newline.

Formatting.