In case CC is set try to split it by spaces into a command and
arguments. This is the same behaviour as in dmd.
Signed-off-by: Andrei Horodniceanu <a.horodniceanu@proton.me>
The .data() method on a StringRef, returns a const char* to the first
character but there is no guarantee that the string is null
terminated.
In order to get a proper null terminated string it is
necessary to use .str() which will return a C++ std::string which
provided the .c_str() method to get the null terminated C string.
The invalid behavior can be seen by running (where the user doesn't
have acces to the inaccessible folder):
$ ldc2 <<< '' - -od=inaccessible/impossible
Error: failed to create output directory: inaccessible/impossible/__stdin_4416.o
The error message should only include inaccessible/impossible.
Signed-off-by: Andrei Horodniceanu <a.horodniceanu@proton.me>
Fix issue where an LLVM error or warning during IR passes (e.g. --fwarn-stack-size) does not abort outputting files. The object file would be added to the cache, and subsequent repeated LDC
execution would fetch the object file from the cache, effectively swallowing the error.
* Improve ftime-trace implementation.
- Rewrite ftime-trace to our own implementatation instead of using LLVM's time trace code. The disadvantage is that this removes LLVM's work from the trace (optimization), but has the large benefit of being able to tailor the tracing output to our needs.
- Add memory tracing to ftime-trace (not possible with LLVM's implementation)
- Do not output the sum for each "category"/named string. This causes the LLVM output to be _very_ long, because we put more information in each time segment name. Tooling that processes the time trace output can do this summing itself (i.e. Tracy), and makes the time trace much more pleasant to view in trace viewers.
- Use MonoTime, move timescale calculation to output stage, 'measurement' stage uses ticks as unit
- Fix crash on `ldc2 -ftime-trace` without files passed.
This uses LLVM's TimeProfiler functionality, such that LLVM's work is traced in the same profile (optimization and machine code gen).
Functionality is meant to be identical to Clang and LLD's --ftime-trace.
And merge the two out-of-sync copies for adding gcc target flags. The flag
was omitted for gcc as linker already, but not for gcc as external assembler.
For invoking MS tools such as link.exe and lib.exe, which require
compatible PATH, LIB, LIBPATH etc.
We previously wrapped the invokation of the tool with a batch file taking
care of setting up the environment and then calling the actual tool.
I changed this to first invoke another batch file setting up its
environment and then dumping all environment variables to a file. LDC then
parses this file and updates its own environment variables. Later spawned
child processes then inherit the updated environment, and the tools will
be located by LLVM in updated PATH.
This allows LDC to keep track of the used Visual C++ installation (and
could so set up additional runtime libs required since VS 2015, something
we would have needed some time back when we still supported VS 2013). It's
also required as a prerequisite for integrating LLD as in-process linker
on Windows.
I quickly looked into the option of setting up the environment manually,
but the MS batch files are full of quirks, mixtures of registry and
file-system lookups, checking for existence of misc. headers to filter out
bogus installations etc. With VS 2017, there's nothing in the registry at
all anymore (at least for the Build Tools); MS expose a COM interface for
querying the VS installer... But at least they still provide the batch
files.
