Rollup merge of #139823 - WaffleLapkin:bootpaper, r=jieyouxu

Fix some bootstrap papercuts

... related to jj and my `./build` symlink setup[^1].

I'm not sure if these are good solutions, but they seem to work. See commits for a bit more info.

r? ``@jieyouxu``

[^1]: see #139804

Author: matthiaskrgr

src/bootstrap/src/core/config/config.rs

@@ -2888,6 +2888,13 @@ impl Config {
 
         let absolute_path = self.src.join(relative_path);
 
+        // NOTE: This check is required because `jj git clone` doesn't create directories for
+        // submodules, they are completely ignored. The code below assumes this directory exists,
+        // so create it here.
+        if !absolute_path.exists() {
+            t!(fs::create_dir_all(&absolute_path));
+        }
+
         // NOTE: The check for the empty directory is here because when running x.py the first time,
         // the submodule won't be checked out. Check it out now so we can build it.
         if !GitInfo::new(false, &absolute_path).is_managed_git_subrepository()

src/tools/compiletest/src/runtest.rs

@@ -2378,12 +2378,19 @@ impl<'test> TestCx<'test> {
         // eg.
         // /home/user/rust/build/x86_64-unknown-linux-gnu/test/ui/<test_dir>/$name.$revision.$mode/
         normalize_path(&self.output_base_dir(), "$TEST_BUILD_DIR");
+        // Same as above, but with a canonicalized path.
+        // This is required because some tests print canonical paths inside test build directory,
+        // so if the build directory is a symlink, normalization doesn't help.
+        //
+        // NOTE: There are also tests which print the non-canonical name, so we need both this and
+        // the above normalizations.
+        normalize_path(&self.output_base_dir().canonicalize_utf8().unwrap(), "$TEST_BUILD_DIR");
         // eg. /home/user/rust/build
         normalize_path(&self.config.build_root, "$BUILD_DIR");
 
         if json {
             // escaped newlines in json strings should be readable
-            // in the stderr files. There's no point int being correct,
+            // in the stderr files. There's no point in being correct,
             // since only humans process the stderr files.
             // Thus we just turn escaped newlines back into newlines.
             normalized = normalized.replace("\\n", "\n");