[clangd] Fix not being able to attach a debugger on macOS

[malaperle-ericsson]

Summary

Clangd is often waiting for input on getline as it awaits requests. If the getline is interrupted, it causes the system call (read) to fail and the EINTR error to be set. This can be seen when attaching a debugger such as LLDB on macOS. On macOS (and possibly other operating systems), this system call is not restarted after interruption but on Linux it is restarted which is why attaching a debugger does work correctly there.

The solution is to work around the non-restarting system call by checking the errno for EINTR when the stream fails and try again. This should be safe on all Unixish platforms.

See also http://bugs.llvm.org/show_bug.cgi?id=32149 for some background discussion.

[bkramer]

Generally makes sense. Is there any reason for the #ifdef? Windows has errno and EINTR too.

[malaperle-ericsson]

In D30675#695370, @bkramer wrote:

Generally makes sense. Is there any reason for the #ifdef? Windows has errno and EINTR too.

I have no reasonable expectation that getline will produce a EINTR on Windows. But perhaps there's no harm in doing so? What is you preference?

[malaperle-ericsson]

Remove ifdexf

[malaperle-ericsson]

Ping :)

[bkramer]

lg, do you have commit access?

[malaperle-ericsson]

In D30675#700917, @bkramer wrote:

lg, do you have commit access?

No I do not have commit access. Could you commit it? Thank you very much!

clangd/ClangDMain.cpp

int main(int argc, char *argv[]) {
   Dispatcher.registerHandler("textDocument/codeAction",
                              llvm::make_unique<CodeActionHandler>(Out, AST));
 
   while (std::cin.good()) {
     // A Language Server Protocol message starts with a HTTP header, delimited
     // by \r\n.
     std::string Line;
     std::getline(std::cin, Line);
+    if (!std::cin.good() && errno == EINTR) {
+      std::cin.clear();
+      continue;
+    }
 
     // Skip empty lines.
     llvm::StringRef LineRef(Line);
     if (LineRef.trim().empty())
       continue;
 
     // We allow YAML-style comments. Technically this isn't part of the
     // LSP specification, but makes writing tests easier.