Compute STDLIB_MODULES with a C auxiliary

Eliminates the dependency on many invocations of `tr` and `cut` to a
single invocation of a C auxiliary.

On Windows, the C auxiliary is considerably faster than the large number
of calls to tr and cut which were necessary in stdlib/StdlibModules.

Changes

@@ -354,11 +354,11 @@ OCaml 4.13.0
   (Gabriel Scherer, review by Nicolás Ojeda Bär, Alain Frisch, Xavier Leroy,
    Daniel Bünzli and Stephen Dolan)
 
-* #10169, #10270, #10301: Use capitalized module names in the Standard Library
-  prefixing scheme to match Dune, e.g. Stdlib__String instead of Stdlib__string.
-  This is a breaking change only to code which attempted to use the internal
-  names before. The Standard Library generated by the Dune rules is now
-  equivalent to the main build (the Dune rules still do not generate a
+* #10169, #10270, #10301, #10451: Use capitalized module names in the Standard
+  Library prefixing scheme to match Dune, e.g. Stdlib__String instead of
+  Stdlib__string. This is a breaking change only to code which attempted to use
+  the internal names before. The Standard Library generated by the Dune rules is
+  now equivalent to the main build (the Dune rules still do not generate a
   distributable compiler).
   (David Allsopp and Mark Shinwell, review by Gabriel Scherer)
 

Makefile

@@ -773,6 +773,9 @@ partialclean::
 
 # The runtime system for the bytecode compiler
 
+$(SAK):
+	$(MAKE) -C runtime sak$(EXE)
+
 .PHONY: runtime
 runtime: stdlib/libcamlrun.$(A)
 

Makefile.common

@@ -160,3 +160,15 @@ OCAMLYACCFLAGS ?=
 
 %.ml %.mli: %.mly
 	$(OCAMLYACC) $(OCAMLYACCFLAGS) $<
+
+SAK = $(ROOTDIR)/runtime/sak$(EXE)
+
+# stdlib/StdlibModules cannot be include'd unless $(SAK) has been built. These
+# two rules add that dependency. They have to be pattern rules since
+# Makefile.common is included before default targets.
+$(ROOTDIR)/%/sak$(EXE):
+	$(MAKE) -C $(ROOTDIR)/$* sak$(EXE)
+
+ifneq "$(REQUIRES_CONFIGURATION)" ""
+$(ROOTDIR)/%/StdlibModules: $(SAK) ;
+endif

api_docgen/Makefile.common

@@ -15,8 +15,8 @@
 ROOTDIR = ..
 DOCGEN= $(ROOTDIR)/api_docgen
 
--include $(ROOTDIR)/stdlib/StdlibModules
 include $(ROOTDIR)/Makefile.common
+include $(ROOTDIR)/stdlib/StdlibModules
 include $(ROOTDIR)/Makefile.best_binaries
 include $(DOCGEN)/Makefile.docfiles
 

manual/tests/Makefile

@@ -1,7 +1,7 @@
 ROOTDIR = ../..
-include $(ROOTDIR)/stdlib/StdlibModules
 include $(ROOTDIR)/api_docgen/Makefile.docfiles
 include $(ROOTDIR)/Makefile.common
+include $(ROOTDIR)/stdlib/StdlibModules
 include $(ROOTDIR)/Makefile.best_binaries
 STDLIBFLAGS = -nostdlib -I $(ROOTDIR)/stdlib
 OCAMLC ?= $(BEST_OCAMLC) $(STDLIBFLAGS)

runtime/Makefile

@@ -142,7 +142,8 @@ endif
 # Build, install and clean targets
 
 .PHONY: all
-all: $(BYTECODE_STATIC_LIBRARIES) $(BYTECODE_SHARED_LIBRARIES) $(PROGRAMS)
+all: $(BYTECODE_STATIC_LIBRARIES) $(BYTECODE_SHARED_LIBRARIES) $(PROGRAMS) \
+     sak$(EXE)
 
 .PHONY: allopt
 ifneq "$(NATIVE_COMPILER)" "false"

runtime/sak.c

@@ -26,6 +26,17 @@
 
 #include <stdio.h>
 #include <string.h>
+#include <ctype.h>
+
+#ifdef _WIN32
+#define strncmp_os wcsncmp
+#define toupper_os towupper
+#define printf_os wprintf
+#else
+#define strncmp_os strncmp
+#define toupper_os toupper
+#define printf_os printf
+#endif
 
 /* Operations
    - encode-C-literal. Used for the OCAML_STDLIB_DIR macro in
@@ -36,6 +47,12 @@
 
      On Windows, `sak encode-C-literal "C:\OCaml🐫\lib"` returns
      `L"C:\\OCaml\xd83d\xdc2b\\lib"`
+   - add-stdlib-prefix. Used in stdlib/StdlibModules to convert the list of
+     basenames given in STDLIB_MODULE_BASENAMES to the actual file basenames
+     in STDLIB_MODULES.
+
+     For example, `sak add-stdlib-prefix stdlib camlinternalAtomic Sys` returns
+     ` stdlib camlinternalAtomic stdlib__Sys`
  */
 
 void usage(void)
@@ -45,6 +62,7 @@ void usage(void)
     "Usage: sak command\n"
     "Commands:\n"
     " * encode-C-literal path - encodes path as a C string literal\n"
+    " * add-stdlib-prefix name1 ... - prefix standard library module names\n"
   );
 }
 
@@ -86,6 +104,31 @@ void encode_C_literal(char_os *path)
   putchar('"');
 }
 
+/* Print the given array of module names to stdout. "stdlib" and names beginning
+   "camlinternal" are printed unaltered. All other names are prefixed "stdlib__"
+   with the original name capitalised (i.e. "foo" prints "stdlib__Foo"). */
+void add_stdlib_prefix(int count, char_os **names)
+{
+  int i;
+  char_os *name;
+
+  for (i = 0; i < count; i++) {
+    name = *names++;
+
+    /* "stdlib" and camlinternal* do not get changed. All other names get
+       capitalised and prefixed "stdlib__". */
+    if (strcmp_os(T("stdlib"), name) == 0
+     || strncmp_os(T("camlinternal"), name, 12) == 0) {
+      printf_os(T(" %s"), name);
+    } else {
+      /* name is a null-terminated string, so an empty string simply has the
+         null-terminator "capitalised". */
+      *name = toupper_os(*name);
+      printf_os(T(" stdlib__%s"), name);
+    }
+  }
+}
+
 #ifdef _WIN32
 int wmain(int argc, wchar_t **argv)
 #else
@@ -94,6 +137,8 @@ int main(int argc, char **argv)
 {
   if (argc == 3 && !strcmp_os(argv[1], T("encode-C-literal"))) {
     encode_C_literal(argv[2]);
+  } else if (argc > 1 && !strcmp_os(argv[1], T("add-stdlib-prefix"))) {
+    add_stdlib_prefix(argc - 2, &argv[2]);
   } else {
     usage();
     return 1;

stdlib/StdlibModules

@@ -15,7 +15,7 @@
 #*                                                                        *
 #**************************************************************************
 
-# This file must be self-contained.
+# This file should be included after Makefile.common
 
 # This file lists all standard library modules. It is used by:
 # 1. stdlib/Makefile when building stdlib.cma
@@ -33,7 +33,7 @@
 
 # Basenames of the source files for the standard library (i.e. unprefixed and
 # with lowercase first letters). These must be listed in dependency order.
-STDLIB_MODULE_BASENAMES=\
+STDLIB_MODULE_BASENAMES = \
   camlinternalFormatBasics camlinternalAtomic \
   stdlib pervasives seq option either result bool char uchar \
   sys list int bytes string unit marshal obj array float int32 int64 nativeint \
@@ -44,20 +44,12 @@ STDLIB_MODULE_BASENAMES=\
   filename complex arrayLabels listLabels bytesLabels stringLabels moreLabels \
   stdLabels bigarray in_channel out_channel
 
-STDLIB_PREFIXED_MODULES=\
+STDLIB_PREFIXED_MODULES = \
   $(filter-out stdlib camlinternal%, $(STDLIB_MODULE_BASENAMES))
 
-define add_stdlib_prefix_first
-$(shell echo $1 | cut -c1 | tr '[:lower:]' '[:upper:]')
-endef
-
-# add stdlib__ as prefix to a module except for internal modules
-# and the stdlib module itself
-define add_stdlib_prefix
-  $(or $(filter-out $(STDLIB_PREFIXED_MODULES), $1), \
-    stdlib__$(call add_stdlib_prefix_first,$1)$(shell echo $1 | cut -c2-))
-endef
-
-STDLIB_MODULES:=\
-  $(foreach module, $(STDLIB_MODULE_BASENAMES), \
-    $(call add_stdlib_prefix,$(module)))
+# The pattern FOO = $(eval FOO := $$(shell <cmd>)$(FOO) ensures that <cmd> is
+# executed either once or not at all, giving us GNU make's equivalent of a
+# string lazy_t.
+STDLIB_MODULES = \
+  $(eval STDLIB_MODULES := $$(shell \
+          $(SAK) add-stdlib-prefix $(STDLIB_MODULE_BASENAMES)))$(STDLIB_MODULES)