[PATCH] D39489: [ELF] - Linkerscript: fix issue with multiple output sections definitions.

[George Rimar via Phabricator via llvm-commits]

grimar added inline comments.


================
Comment at: ELF/ScriptParser.cpp:974
     StringRef Name = readParenLiteral();
-    OutputSection *Sec = Script->getOrCreateOutputSection(Name);
+    Script->declareOutputSection(Name);
     return [=]() -> ExprValue {
----------------
peter.smith wrote:
> If I understand correctly this might create a forward reference to some OutputSection Name that we've not seen yet? I don't think that this is allowed in this case: https://sourceware.org/binutils/docs/ld/Builtin-Functions.html
> > Return the address (VMA) of the named section. Your script must previously have defined the location of that section. 
> 
It is actually what original code did already before my change. `getOrCreateOutputSection` creates forward reference.
We rely on that in our testcases, for example absolute.s has just:
"PROVIDE(foo = 1 + ABSOLUTE(ADDR(.text)));" and expects we can evaluate that. I believe we have the same
behavior as gnu linkers here.

Below you referenced to spec saying that for ALIGNOF "If the section *has not been allocated* when this is evaluated, the linker will report an error.". I think that is how it should say for all commands you mentioned.
We use `checkIfExists` to verify that section was allocated at the moment when we call evaluation of `ADDR` or other commands, and for parsing script we create forward references atm.






https://reviews.llvm.org/D39489