[cfe-dev] [libcxx-dev] Is the C++ filesystem object file installed via LLVM's APT repository missing symbols?

[Michael Spencer via cfe-dev]

On Tue, Apr 2, 2019 at 3:49 PM Brian Cain via cfe-dev <
cfe-dev at lists.llvm.org> wrote:

> This is the intended design of GNU ld.  Some linkers follow this behavior,
> some don’t.  e.g. IIRC lld chose to have an implicit grouping (like ld’s
> start-group/end-group), which comes w/pros and cons.  Pros: like you said,
> you needn’t concern yourself with the order of the libs.  If you are
> consuming several different libraries, one could see the perspective that
> from up here at the executable’s perspective, you don’t care how the linker
> does symbol resolution and one could imagine that it might not even matter
> that it takes more time for it to go and search the ones that appear
> earlier on the command line.  Cons: increased link time, cyclical
> dependencies among libraries are easy to introduce now because the linker
> won’t alert you to the problem.
>

This doesn't really have an impact on link times in lld (note that lld is
faster than both gnu ld and gold).  This is because lld just adds all
symbols to its internal symbol table, but unresolved archive symbols get
added as lazy symbols that don't show up in the output unless they are
actually referenced.  It's slower in other linkers because they do another
full iteration over the symbol table every time an archive is visited.  The
biggest issue I saw with this is that you actually have different semantics
for what gets linked, but it turned out that every case where this broke
the link was really a bug in the program being linked, often duplicate
definitions of stuff.

- Michael Spencer


>
>
> Relevant details from the ld.bfd manpage:
>
>
>
> --start-group archives --end-group
>
>
>
> …
>
> The specified archives are searched repeatedly until no new undefined
> references are created. *Normally, an archive is searched only once in
> the order that it is specified on the command line. If a symbol in that
> archive is needed to resolve an undefined symbol referred to by an object
> in an archive that appears later on the command line, the linker would not
> be able to resolve that reference.* By grouping the archives, they all be
> searched repeatedly until all possible references are resolved.
>
> Using this option has a significant performance cost. It is best to use it
> only when there are unavoidable circular references between two or more
> archives.
>
>
>
> *From:* libcxx-dev <libcxx-dev-bounces at lists.llvm.org> *On Behalf Of *Louis
> Dionne via libcxx-dev
>
> To clarify, what I meant is that from a very naive point of view, the
> order in which you provide `-l`'s on the command line shouldn't matter --
> it should work as long as you provide the right set of `-l`'s. I'm puzzled
> every time a tool (e.g. `ld`) behaves differently based on the order of
> command line arguments.
>
>
>
> Maybe there's a good reason for this, though. Maybe the cfe-dev folks can
> explain?
>
>
>
> Louis
>
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20190403/a57ffe5c/attachment.html>