Aug 25 2018
Aug 23 2018
Thanks for catching this! On top of respecting our API contract, it's good to be safe around C strings.
May 21 2018
@aprantl Let me know if this is the right way to address that FIXME. There didn't seem to be any consequences of the small change I've got here which could be very good or very bad from my naive perspective.
May 20 2018
May 19 2018
May 18 2018
First, the naming here is absurdly confusing. Second, it's also inconsistent with all other functions, e.g. LLVMGetSourceFileName/LLVMSetSourceFileName.
Use LLVMGetValueName2 and LLVMSetValueName2, since this is what we do when updating function signatures in the C API.
May 15 2018
May 14 2018
May 13 2018
May 12 2018
May 11 2018
May 10 2018
You could write the same tests as in llvm-c-test/echo.cpp. But I would not block merging on this.
May 9 2018
llvm::Value::replaceAllUsesWith and llvm::MDNode::replaceAllUsesWith are distinct operations. The former assumes the source and replacement values are the same type, the latter assumes that you're trying to build up a graph of dummy metadata for an incomplete or recursive definition that you will eventually finalize by replacing it with real nodes.
May 8 2018
Apr 28 2018
Apr 27 2018
Unless there has been a release since 6.0(.1), we should be free to change these. We merged the referenced patch after that.
I've only seen it used to discriminate between different kinds of DIScopes. If it's really an implementation detail, I can abandon this patch.
Apr 26 2018
Apr 23 2018
A note from the test: I removed foo_inner_scope because I no longer need it to anchor the lexical block scope now that I can declare variables.
Apr 22 2018
This is expected to conflict with D45933. Whichever lands first, I'll rebase the other.
Apr 10 2018
Going to commit this immediately to fix the bots.
Apr 7 2018
Apr 6 2018
Apr 5 2018
I'll need to move the Function scope API as well to write a coherent test.
The original API for LLVMSetModuleInlineAsm implies that only C-style strings can be passed here, but the implementation is copying to a std::string under the hood anyhow. If that's the invariant that we want to maintain here, I will revert the deprecation and add documentation.
Mar 30 2018
Mar 28 2018
Respond to review nits