[AVR][clang] Search for avr-libc in $SYSROOT/avr

[mhjacobson]

Summary

The paths in PossibleAVRLibcLocations are appended with /lib/ and /include/ to find the binaries and headers, respectively, for avr-libc.

It's common for avr-libc to live at $SYSROOT/avr/{lib,include}/, e.g. when $SYSROOT is, for example, /opt/local/.

[mhjacobson]

I'm also tempted to remove /usr/lib/avr, because these paths make no sense to me: $SYSROOT/usr/lib/avr/lib and $SYSROOT/usr/lib/avr/include.

But I haven't done that here, in case there's some reason those paths do make sense. (Also, the files in clang/test/Driver/Inputs/basic_avr_tree make use of the /usr/lib/avr/{lib,include} structure.)

[mhjacobson]

In D107672#2932190, @mhjacobson wrote:

I'm also tempted to remove /usr/lib/avr, because these paths make no sense to me: $SYSROOT/usr/lib/avr/lib and $SYSROOT/usr/lib/avr/include.

Ah. According to https://reviews.llvm.org/D54334, that was added because that's where Ubuntu installs avr-libc. OK then!

[MaskRay]

I know nearly nothing about AVR, but from avr-gcc a.c '-###' I doubt this is the GCC behavior. GCC's library path is relative to the GCC installation. /usr/lib/gcc/avr/5.4.0/../../../avr/lib

There is a difference between sysroot inferred path and GCC installation inferred path. https://maskray.me/blog/2021-03-28-compiler-driver-and-cross-compilation

[mhjacobson]

In D107672#2932406, @MaskRay wrote:

I know nearly nothing about AVR, but from avr-gcc a.c '-###' I doubt this is the GCC behavior. GCC's library path is relative to the GCC installation. /usr/lib/gcc/avr/5.4.0/../../../avr/lib

There is a difference between sysroot inferred path and GCC installation inferred path. https://maskray.me/blog/2021-03-28-compiler-driver-and-cross-compilation

Wow, nice write-up! I'll have to spend some more time to digest it fully.

At first glance, I think you're right: avr-gcc does expect avr-libc to be relative to itself. That seems slightly weird to me (since avr-libc is technically separate from GCC), but OK.

clang's driver certainly searches for libgcc and avr-libc separately. I guess you're saying it shouldn't?

Obviously the separate searching isn't something introduced here, so (unless you object) I'll file a bug report about it.

[MaskRay]

@benshi001 This was prematurely committed. It has no test and has apparent unaddressed issue. (I revamped many parts in the Linux/Gnu search paths so am quite confident about my observation.)

[mhjacobson]

In D107672#2932413, @mhjacobson wrote:

At first glance, I think you're right: avr-gcc does expect avr-libc to be relative to itself. That seems slightly weird to me (since avr-libc is technically separate from GCC), but OK.

OK, after looking at the GCC source, I see there's nothing specific to avr-libc. It's just that GCC basically tries to look in $PREFIX/[machine/]lib, where $PREFIX is actually the path to GCC, dot-dotted up to where it thinks the prefix lives, to allow for the possibility that the whole prefix was relocated.

So, @MaskRay, correct me if I'm wrong: it seems like you're suggesting that we use something derived from the GCCInstallationDetector to find libc? Is there another platform's driver I can look at as an example?

[mhjacobson]

In D107672#2932423, @MaskRay wrote:

@benshi001 This was prematurely committed. It has no test and has apparent unaddressed issue. (I revamped many parts in the Linux/Gnu search paths so am quite confident about my observation.)

I'm working on a test now, and I'll submit a follow-up patch adding it, if that's OK.

[benshi001]

In D107672#2932442, @mhjacobson wrote:

In D107672#2932413, @mhjacobson wrote:

At first glance, I think you're right: avr-gcc does expect avr-libc to be relative to itself. That seems slightly weird to me (since avr-libc is technically separate from GCC), but OK.

OK, after looking at the GCC source, I see there's nothing specific to avr-libc. It's just that GCC basically tries to look in $PREFIX/[machine/]lib, where $PREFIX is actually the path to GCC, dot-dotted up to where it thinks the prefix lives, to allow for the possibility that the whole prefix was relocated.

So, @MaskRay, correct me if I'm wrong: it seems like you're suggesting that we use something derived from the GCCInstallationDetector to find libc? Is there another platform's driver I can look at as an example?

I think so. It would be better to search avr-libc relatively from avr-gcc/../../../, and have submit it as https://reviews.llvm.org/D107682.

@mhjacobson, could you please check if it work on your platform? It does work on my ubuntu system.

clang/lib/Driver/ToolChains/AVR.cpp

 llvm::Optional<unsigned> GetMCUSectionAddressData(StringRef MCUName) {
   for (const auto &MCU : MCUInfo)
     if (MCU.Name == MCUName && MCU.DataAddr > 0)
       return Optional<unsigned>(MCU.DataAddr);
   return Optional<unsigned>();
 }
 
 const StringRef PossibleAVRLibcLocations[] = {
+    "/avr",
     "/usr/avr",
     "/usr/lib/avr",
 };
 
 } // end anonymous namespace
 
 /// AVR Toolchain
 AVRToolChain::AVRToolChain(const Driver &D, const llvm::Triple &Triple,