External package detection for compilers (#43464)

This creates shared infrastructure for compiler packages to implement the 
detailed search capabilities from the `spack compiler find` command for the 
`spack external find` command.

After this commit, `spack compiler find` can be replaced with 
`spack external find --tag compiler`, with the exception of mixed toolchains.

lib/spack/docs/getting_started.rst

@@ -478,6 +478,13 @@ prefix, you can add them to the ``extra_attributes`` field. Similarly,
 all other fields from the compilers config can be added to the
 ``extra_attributes`` field for an external representing a compiler.
 
+Note that the format for the ``paths`` field in the
+``extra_attributes`` section is different than in the ``compilers``
+config. For compilers configured as external packages, the section is
+named ``compilers`` and the dictionary maps language names (``c``,
+``cxx``, ``fortran``) to paths, rather than using the names ``cc``,
+``fc``, and ``f77``.
+
 .. code-block:: yaml
 
    packages:
@@ -493,11 +500,10 @@ all other fields from the compilers config can be added to the
        - spec: llvm+clang@15.0.0 arch=linux-rhel8-skylake
          prefix: /usr
          extra_attributes:
-           paths:
-             cc: /usr/bin/clang-with-suffix
+           compilers:
+             c: /usr/bin/clang-with-suffix
              cxx: /usr/bin/clang++-with-extra-info
-             fc: /usr/bin/gfortran
-             f77: /usr/bin/gfortran
+             fortran: /usr/bin/gfortran
            extra_rpaths:
            - /usr/lib/llvm/
 

lib/spack/spack/build_systems/compiler.py

@@ -0,0 +1,144 @@
+# Copyright 2013-2023 Lawrence Livermore National Security, LLC and other
+# Spack Project Developers. See the top-level COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+import itertools
+import os
+import pathlib
+import re
+import sys
+from typing import Dict, List, Sequence, Tuple, Union
+
+import llnl.util.tty as tty
+from llnl.util.lang import classproperty
+
+import spack.compiler
+import spack.package_base
+
+# Local "type" for type hints
+Path = Union[str, pathlib.Path]
+
+
+class CompilerPackage(spack.package_base.PackageBase):
+    """A Package mixin for all common logic for packages that implement compilers"""
+
+    # TODO: how do these play nicely with other tags
+    tags: Sequence[str] = ["compiler"]
+
+    #: Optional suffix regexes for searching for this type of compiler.
+    #: Suffixes are used by some frameworks, e.g. macports uses an '-mp-X.Y'
+    #: version suffix for gcc.
+    compiler_suffixes: List[str] = [r"-.*"]
+
+    #: Optional prefix regexes for searching for this compiler
+    compiler_prefixes: List[str] = []
+
+    #: Compiler argument(s) that produces version information
+    #: If multiple arguments, the earlier arguments must produce errors when invalid
+    compiler_version_argument: Union[str, Tuple[str]] = "-dumpversion"
+
+    #: Regex used to extract version from compiler's output
+    compiler_version_regex: str = "(.*)"
+
+    #: Static definition of languages supported by this class
+    compiler_languages: Sequence[str] = ["c", "cxx", "fortran"]
+
+    def __init__(self, spec: "spack.spec.Spec"):
+        super().__init__(spec)
+        msg = f"Supported languages for {spec} are not a subset of possible supported languages"
+        msg += f"    supports: {self.supported_languages}, valid values: {self.compiler_languages}"
+        assert set(self.supported_languages) <= set(self.compiler_languages), msg
+
+    @property
+    def supported_languages(self) -> Sequence[str]:
+        """Dynamic definition of languages supported by this package"""
+        return self.compiler_languages
+
+    @classproperty
+    def compiler_names(cls) -> Sequence[str]:
+        """Construct list of compiler names from per-language names"""
+        names = []
+        for language in cls.compiler_languages:
+            names.extend(getattr(cls, f"{language}_names"))
+        return names
+
+    @classproperty
+    def executables(cls) -> Sequence[str]:
+        """Construct executables for external detection from names, prefixes, and suffixes."""
+        regexp_fmt = r"^({0}){1}({2})$"
+        prefixes = [""] + cls.compiler_prefixes
+        suffixes = [""] + cls.compiler_suffixes
+        if sys.platform == "win32":
+            ext = r"\.(?:exe|bat)"
+            suffixes += [suf + ext for suf in suffixes]
+        return [
+            regexp_fmt.format(prefix, re.escape(name), suffix)
+            for prefix, name, suffix in itertools.product(prefixes, cls.compiler_names, suffixes)
+        ]
+
+    @classmethod
+    def determine_version(cls, exe: Path):
+        version_argument = cls.compiler_version_argument
+        if isinstance(version_argument, str):
+            version_argument = (version_argument,)
+
+        for va in version_argument:
+            try:
+                output = spack.compiler.get_compiler_version_output(exe, va)
+                match = re.search(cls.compiler_version_regex, output)
+                if match:
+                    return ".".join(match.groups())
+            except spack.util.executable.ProcessError:
+                pass
+            except Exception as e:
+                tty.debug(
+                    f"[{__file__}] Cannot detect a valid version for the executable "
+                    f"{str(exe)}, for package '{cls.name}': {e}"
+                )
+
+    @classmethod
+    def compiler_bindir(cls, prefix: Path) -> Path:
+        """Overridable method for the location of the compiler bindir within the preifx"""
+        return os.path.join(prefix, "bin")
+
+    @classmethod
+    def determine_compiler_paths(cls, exes: Sequence[Path]) -> Dict[str, Path]:
+        """Compute the paths to compiler executables associated with this package
+
+        This is a helper method for ``determine_variants`` to compute the ``extra_attributes``
+        to include with each spec object."""
+        # There are often at least two copies (not symlinks) of each compiler executable in the
+        # same directory: one with a canonical name, e.g. "gfortran", and another one with the
+        # target prefix, e.g. "x86_64-pc-linux-gnu-gfortran". There also might be a copy of "gcc"
+        # with the version suffix, e.g. "x86_64-pc-linux-gnu-gcc-6.3.0". To ensure the consistency
+        # of values in the "paths" dictionary (i.e. we prefer all of them to reference copies
+        # with canonical names if possible), we iterate over the executables in the reversed sorted
+        # order:
+        # First pass over languages identifies exes that are perfect matches for canonical names
+        # Second pass checks for names with prefix/suffix
+        # Second pass is sorted by language name length because longer named languages
+        # e.g. cxx can often contain the names of shorter named languages
+        # e.g. c (e.g. clang/clang++)
+        paths = {}
+        exes = sorted(exes, reverse=True)
+        languages = {
+            lang: getattr(cls, f"{lang}_names")
+            for lang in sorted(cls.compiler_languages, key=len, reverse=True)
+        }
+        for exe in exes:
+            for lang, names in languages.items():
+                if os.path.basename(exe) in names:
+                    paths[lang] = exe
+                    break
+            else:
+                for lang, names in languages.items():
+                    if any(name in os.path.basename(exe) for name in names):
+                        paths[lang] = exe
+                        break
+
+        return paths
+
+    @classmethod
+    def determine_variants(cls, exes: Sequence[Path], version_str: str) -> Tuple:
+        # path determination is separated so it can be reused in subclasses
+        return "", {"compilers": cls.determine_compiler_paths(exes=exes)}

lib/spack/spack/compilers/__init__.py

@@ -164,33 +164,56 @@ def _compiler_config_from_package_config(config):
 
 
 def _compiler_config_from_external(config):
+    extra_attributes_key = "extra_attributes"
+    compilers_key = "compilers"
+    c_key, cxx_key, fortran_key = "c", "cxx", "fortran"
+
+    # Allow `@x.y.z` instead of `@=x.y.z`
     spec = spack.spec.parse_with_version_concrete(config["spec"])
-    # use str(spec.versions) to allow `@x.y.z` instead of `@=x.y.z`
+
     compiler_spec = spack.spec.CompilerSpec(
         package_name_to_compiler_name.get(spec.name, spec.name), spec.version
     )
 
-    extra_attributes = config.get("extra_attributes", {})
-    prefix = config.get("prefix", None)
+    err_header = f"The external spec '{spec}' cannot be used as a compiler"
 
-    compiler_class = class_for_compiler_name(compiler_spec.name)
-    paths = extra_attributes.get("paths", {})
-    compiler_langs = ["cc", "cxx", "fc", "f77"]
-    for lang in compiler_langs:
-        if paths.setdefault(lang, None):
-            continue
-
-        if not prefix:
-            continue
-
-        # Check for files that satisfy the naming scheme for this compiler
-        bindir = os.path.join(prefix, "bin")
-        for f, regex in itertools.product(os.listdir(bindir), compiler_class.search_regexps(lang)):
-            if regex.match(f):
-                paths[lang] = os.path.join(bindir, f)
-
-    if all(v is None for v in paths.values()):
+    # If extra_attributes is not there I might not want to use this entry as a compiler,
+    # therefore just leave a debug message, but don't be loud with a warning.
+    if extra_attributes_key not in config:
+        tty.debug(f"[{__file__}] {err_header}: missing the '{extra_attributes_key}' key")
         return None
+    extra_attributes = config[extra_attributes_key]
+
+    # If I have 'extra_attributes' warn if 'compilers' is missing, or we don't have a C compiler
+    if compilers_key not in extra_attributes:
+        warnings.warn(
+            f"{err_header}: missing the '{compilers_key}' key under '{extra_attributes_key}'"
+        )
+        return None
+    attribute_compilers = extra_attributes[compilers_key]
+
+    if c_key not in attribute_compilers:
+        warnings.warn(
+            f"{err_header}: missing the C compiler path under "
+            f"'{extra_attributes_key}:{compilers_key}'"
+        )
+        return None
+    c_compiler = attribute_compilers[c_key]
+
+    # C++ and Fortran compilers are not mandatory, so let's just leave a debug trace
+    if cxx_key not in attribute_compilers:
+        tty.debug(f"[{__file__}] The external spec {spec} does not have a C++ compiler")
+
+    if fortran_key not in attribute_compilers:
+        tty.debug(f"[{__file__}] The external spec {spec} does not have a Fortran compiler")
+
+    # compilers format has cc/fc/f77, externals format has "c/fortran"
+    paths = {
+        "cc": c_compiler,
+        "cxx": attribute_compilers.get(cxx_key, None),
+        "fc": attribute_compilers.get(fortran_key, None),
+        "f77": attribute_compilers.get(fortran_key, None),
+    }
 
     if not spec.architecture:
         host_platform = spack.platforms.host()

lib/spack/spack/package.py

@@ -39,6 +39,7 @@
 )
 from spack.build_systems.cargo import CargoPackage
 from spack.build_systems.cmake import CMakePackage, generator
+from spack.build_systems.compiler import CompilerPackage
 from spack.build_systems.cuda import CudaPackage
 from spack.build_systems.generic import Package
 from spack.build_systems.gnu import GNUMirrorPackage

lib/spack/spack/test/cmd/compiler.py

@@ -261,15 +261,14 @@ def test_compiler_list_empty(no_compilers_yaml, working_env, compilers_dir):
     [
         (
             {
-                "spec": "gcc@=7.7.7 os=foobar target=x86_64",
+                "spec": "gcc@=7.7.7 languages=c,cxx,fortran os=foobar target=x86_64",
                 "prefix": "/path/to/fake",
                 "modules": ["gcc/7.7.7", "foobar"],
                 "extra_attributes": {
-                    "paths": {
-                        "cc": "/path/to/fake/gcc",
+                    "compilers": {
+                        "c": "/path/to/fake/gcc",
                         "cxx": "/path/to/fake/g++",
-                        "fc": "/path/to/fake/gfortran",
-                        "f77": "/path/to/fake/gfortran",
+                        "fortran": "/path/to/fake/gfortran",
                     },
                     "flags": {"fflags": "-ffree-form"},
                 },
@@ -285,26 +284,7 @@ def test_compiler_list_empty(no_compilers_yaml, working_env, compilers_dir):
 \tmodules  = ['gcc/7.7.7', 'foobar']
 \toperating system  = foobar
 """,
-        ),
-        (
-            {
-                "spec": "gcc@7.7.7",
-                "prefix": "{prefix}",
-                "modules": ["gcc/7.7.7", "foobar"],
-                "extra_attributes": {"flags": {"fflags": "-ffree-form"}},
-            },
-            """gcc@7.7.7:
-\tpaths:
-\t\tcc = {compilers_dir}{sep}gcc-8{suffix}
-\t\tcxx = {compilers_dir}{sep}g++-8{suffix}
-\t\tf77 = {compilers_dir}{sep}gfortran-8{suffix}
-\t\tfc = {compilers_dir}{sep}gfortran-8{suffix}
-\tflags:
-\t\tfflags = ['-ffree-form']
-\tmodules  = ['gcc/7.7.7', 'foobar']
-\toperating system  = debian6
-""",
-        ),
+        )
     ],
 )
 def test_compilers_shows_packages_yaml(

lib/spack/spack/test/conftest.py

@@ -1694,7 +1694,7 @@ def mock_executable(tmp_path):
     """Factory to create a mock executable in a temporary directory that
     output a custom string when run.
     """
-    shebang = "#!/bin/sh\n" if sys.platform != "win32" else "@ECHO OFF"
+    shebang = "#!/bin/sh\n" if sys.platform != "win32" else "@ECHO OFF\n"
 
     def _factory(name, output, subdir=("bin",)):
         executable_dir = tmp_path.joinpath(*subdir)