build: Make genlast write the "split" files

This gets a further speedup of about 2s (12s -> 9.5s elapsed build time)
for stm32f405_feather

For what are probably historical reasons, the qstr process involves
preprocessing a large number of source files into a single "qstr.i.last"
file, then reading this and splitting it into one "qstr" file for each
original source ("*.c") file.

By eliminating the step of writing qstr.i.last as well as making the
regular-expression-matching part be parallelized, build speed is further
improved.

Because the step to build QSTR_DEFS_COLLECTED does not access
qstr.i.last, the path is replaced with "-" in the Makefile.

2 files changed, 58 insertions, 29 deletions

diff --git a/py/genlast.py b/py/genlast.py
index e14e7ef1e..1df2a2482 100644
--- a/py/genlast.py
+++ b/py/genlast.py
@@ -1,38 +1,71 @@
 #!/usr/bin/env python3
 
 import sys
-from concurrent.futures import ThreadPoolExecutor
+import re
+import os
+import itertools
+from concurrent.futures import ProcessPoolExecutor
 import multiprocessing
 import threading
 import subprocess
 
+from makeqstrdefs import qstr_unescape, QSTRING_BLACK_LIST
+
+re_line = re.compile(r"#[line]*\s(\d+)\s\"([^\"]+)\"", re.DOTALL)
+re_qstr = re.compile(r'MP_QSTR_[_a-zA-Z0-9]+', re.DOTALL)
+re_translate = re.compile(r'translate\(\"((?:(?=(\\?))\2.)*?)\"\)', re.DOTALL)
+
+def write_out(fname, output_dir, output):
+    if output:
+        for m, r in [("/", "__"), ("\\", "__"), (":", "@"), ("..", "@@")]:
+            fname = fname.replace(m, r)
+        with open(output_dir + "/" + fname + ".qstr", "w") as f:
+            f.write("\n".join(output) + "\n")
+
+def process_file(fname, output_dir, content):
+    content = content.decode('utf-8', errors='ignore')
+    output = []
+    for match in re_qstr.findall(content):
+        name = match.replace('MP_QSTR_', '')
+        if name not in QSTRING_BLACK_LIST:
+            output.append('Q(' + qstr_unescape(name) + ')')
+    for match in re_translate.findall(content):
+        output.append('TRANSLATE("' + match[0] + '")')
+
+    write_out(fname, output_dir, output)
+
 
 def checkoutput1(args):
     info = subprocess.run(args, check=True, stdout=subprocess.PIPE, input='')
     return info.stdout
 
-idx1 = sys.argv.index('--')
-idx2 = sys.argv.index('--', idx1+1)
-check = sys.argv[1:idx1]
-always = sys.argv[idx1+1:idx2]
-command = sys.argv[idx2+1:]
-
-output_lock = threading.Lock()
-def preprocess(fn):
-    output = checkoutput1(command + [fn])
-    # Ensure our output doesn't interleave with others
-    # a threading.Lock is not a context manager object :(
+def preprocess(command, output_dir, fn):
     try:
-        output_lock.acquire()
-        sys.stdout.buffer.write(output)
-    finally:
-        output_lock.release()
+        output = checkoutput1(command + [fn])
+        process_file(fn, output_dir, output)
+    except Exception as e:
+        print(e, file=sys.stderr)
 
-def maybe_preprocess(fn):
+def maybe_preprocess(command, output_dir, fn):
     if subprocess.call(["grep", "-lqE", "(MP_QSTR|translate)", fn]) == 0:
-        preprocess(fn)
+        preprocess(command, output_dir, fn)
+
+if __name__ == '__main__':
+
+
+    idx1 = sys.argv.index('--')
+    idx2 = sys.argv.index('--', idx1+1)
+    output_dir = sys.argv[1]
+    check = sys.argv[2:idx1]
+    always = sys.argv[idx1+1:idx2]
+    command = sys.argv[idx2+1:]
+
+    if not os.path.isdir(output_dir):
+        os.makedirs(output_dir)
 
-executor = ThreadPoolExecutor(max_workers=multiprocessing.cpu_count() + 1)
-executor.map(maybe_preprocess, check)
-executor.map(preprocess, always)
-executor.shutdown()
+    # Mac and Windows use 'spawn'.  Uncomment this during testing to catch spawn-specific problems on Linux.
+    #multiprocessing.set_start_method("spawn")
+    executor = ProcessPoolExecutor(max_workers=multiprocessing.cpu_count() + 1)
+    executor.map(maybe_preprocess, itertools.repeat(command), itertools.repeat(output_dir), check)
+    executor.map(preprocess, itertools.repeat(command), itertools.repeat(output_dir), always)
+    executor.shutdown()
diff --git a/py/mkrules.mk b/py/mkrules.mk
index 84e16ec96..549b0e5e4 100644
--- a/py/mkrules.mk
+++ b/py/mkrules.mk
@@ -76,18 +76,14 @@ $(OBJ): | $(HEADER_BUILD)/qstrdefs.enum.h $(HEADER_BUILD)/mpversion.h
 # - if anything in QSTR_GLOBAL_DEPENDENCIES is newer, then process all source files ($^)
 # - else, if list of newer prerequisites ($?) is not empty, then process just these ($?)
 # - else, process all source files ($^) [this covers "make -B" which can set $? to empty]
-$(HEADER_BUILD)/qstr.i.last: $(SRC_QSTR) $(SRC_QSTR_PREPROCESSOR) $(QSTR_GLOBAL_DEPENDENCIES) | $(HEADER_BUILD)/mpversion.h
+$(HEADER_BUILD)/qstr.split: $(SRC_QSTR) $(SRC_QSTR_PREPROCESSOR) $(QSTR_GLOBAL_DEPENDENCIES) | $(HEADER_BUILD)/mpversion.h $(PY_SRC)/genlast.py
 	$(STEPECHO) "GEN $@"
-	$(Q)$(PYTHON3) $(PY_SRC)/genlast.py $(if $(filter $?,$(QSTR_GLOBAL_DEPENDENCIES)),$^,$(if $?,$?,$^)) --  $(SRC_QSTR_PREPROCESSOR) -- $(CPP) $(QSTR_GEN_EXTRA_CFLAGS) $(CFLAGS) >$(HEADER_BUILD)/qstr.i.last;
-
-$(HEADER_BUILD)/qstr.split: $(HEADER_BUILD)/qstr.i.last $(PY_SRC)/makeqstrdefs.py
-	$(STEPECHO) "GEN $@"
-	$(Q)$(PYTHON3) $(PY_SRC)/makeqstrdefs.py split $(HEADER_BUILD)/qstr.i.last $(HEADER_BUILD)/qstr $(QSTR_DEFS_COLLECTED)
+	$(Q)$(PYTHON3) $(PY_SRC)/genlast.py $(HEADER_BUILD)/qstr $(if $(filter $?,$(QSTR_GLOBAL_DEPENDENCIES)),$^,$(if $?,$?,$^)) --  $(SRC_QSTR_PREPROCESSOR) -- $(CPP) $(QSTR_GEN_EXTRA_CFLAGS) $(CFLAGS)
 	$(Q)touch $@
 
 $(QSTR_DEFS_COLLECTED): $(HEADER_BUILD)/qstr.split $(PY_SRC)/makeqstrdefs.py
 	$(STEPECHO) "GEN $@"
-	$(Q)$(PYTHON3) $(PY_SRC)/makeqstrdefs.py cat $(HEADER_BUILD)/qstr.i.last $(HEADER_BUILD)/qstr $(QSTR_DEFS_COLLECTED)
+	$(Q)$(PYTHON3) $(PY_SRC)/makeqstrdefs.py cat - $(HEADER_BUILD)/qstr $(QSTR_DEFS_COLLECTED)
 
 # $(sort $(var)) removes duplicates
 #