[mypyc] Support overriding the group name used in output files (#19272)

JukkaL · web-flow · commit 75f7a2579179 · 2025-06-11T11:22:54.000+01:00
By default, when compiling more than one file, a shared library
is generated with a file name derived from the sha of compiled modules
(i.e. the group name). This is also used in the names of generated
.c and .h files. Add experimental `group_name` argument to `mypycify`
that allows overriding this. This can be useful when integrating
mypyc to a build system, as this makes the names of output files
more predictable.
diff --git a/mypyc/build.py b/mypyc/build.py
@@ -357,6 +357,7 @@ def construct_groups(
     sources: list[BuildSource],
     separate: bool | list[tuple[list[str], str | None]],
     use_shared_lib: bool,
+    group_name_override: str | None,
 ) -> emitmodule.Groups:
     """Compute Groups given the input source list and separate configs.
 
@@ -386,7 +387,10 @@ def construct_groups(
     # Generate missing names
     for i, (group, name) in enumerate(groups):
         if use_shared_lib and not name:
-            name = group_name([source.module for source in group])
+            if group_name_override is not None:
+                name = group_name_override
+            else:
+                name = group_name([source.module for source in group])
         groups[i] = (group, name)
 
     return groups
@@ -432,7 +436,10 @@ def mypyc_build(
         or always_use_shared_lib
     )
 
-    groups = construct_groups(mypyc_sources, separate, use_shared_lib)
+    groups = construct_groups(mypyc_sources, separate, use_shared_lib, compiler_options.group_name)
+
+    if compiler_options.group_name is not None:
+        assert len(groups) == 1, "If using custom group_name, only one group is expected"
 
     # We let the test harness just pass in the c file contents instead
     # so that it can do a corner-cutting version without full stubs.
@@ -477,6 +484,7 @@ def mypycify(
     target_dir: str | None = None,
     include_runtime_files: bool | None = None,
     strict_dunder_typing: bool = False,
+    group_name: str | None = None,
 ) -> list[Extension]:
     """Main entry point to building using mypyc.
 
@@ -519,6 +527,10 @@ def mypycify(
         strict_dunder_typing: If True, force dunder methods to have the return type
                               of the method strictly, which can lead to more
                               optimization opportunities. Defaults to False.
+        group_name: If set, override the default group name derived from
+                    the hash of module names. This is used for the names of the
+                    output C files and the shared library. This is only supported
+                    if there is a single group. [Experimental]
     """
 
     # Figure out our configuration
@@ -530,6 +542,7 @@ def mypycify(
         target_dir=target_dir,
         include_runtime_files=include_runtime_files,
         strict_dunder_typing=strict_dunder_typing,
+        group_name=group_name,
     )
 
     # Generate all the actual important C code
diff --git a/mypyc/options.py b/mypyc/options.py
@@ -15,6 +15,7 @@ def __init__(
         capi_version: tuple[int, int] | None = None,
         python_version: tuple[int, int] | None = None,
         strict_dunder_typing: bool = False,
+        group_name: str | None = None,
     ) -> None:
         self.strip_asserts = strip_asserts
         self.multi_file = multi_file
@@ -38,3 +39,9 @@ def __init__(
         # will assume the return type of the method strictly, which can lead to
         # more optimization opportunities.
         self.strict_dunders_typing = strict_dunder_typing
+        # Override the automatic group name derived from the hash of module names.
+        # This affects the names of generated .c, .h and shared library files.
+        # This is only supported when compiling exactly one group, and a shared
+        # library is generated (with shims). This can be used to make the output
+        # file names more predictable.
+        self.group_name = group_name
diff --git a/mypyc/test/test_run.py b/mypyc/test/test_run.py
@@ -235,7 +235,7 @@ def run_case_step(self, testcase: DataDrivenTestCase, incremental_step: int) ->
             else False
         )
 
-        groups = construct_groups(sources, separate, len(module_names) > 1)
+        groups = construct_groups(sources, separate, len(module_names) > 1, None)
 
         try:
             compiler_options = CompilerOptions(

Original file line number	Diff line number	Diff line change
`@@ -235,7 +235,7 @@ def run_case_step(self, testcase: DataDrivenTestCase, incremental_step: int) ->`
`235`	`235`	`else False`
`236`	`236`	`)`
`237`	`237`
`238`		`- groups = construct_groups(sources, separate, len(module_names) > 1)`
	`238`	`+ groups = construct_groups(sources, separate, len(module_names) > 1, None)`
`239`	`239`
`240`	`240`	`try:`
`241`	`241`	`compiler_options = CompilerOptions(`