On 2/24/24 4:47 PM, Collin Funk wrote:
> Anyways, upon further inspection not all of the gnulib-modules are
> sorted in merge-gnulib. When "unlocked-io" was added to Emacs it was
> placed after "update-copyright" [1]. I assume that they are sorted
> somewhere before the actioncmd step in gnulib-tool. Let me experiment
> with the --avoid modules and I'll reply if I notice anything.
I think that I have confirmed this behavior by changing the order of
"GNULIB_MODULES" and "AVOIDED_MODULES" in Emacs admin/merge-gnulib
script. Changing GNULIB_MODULES so the first modules are ordered 1. dup2,
2. alignasof, 3. copy-file-range, 4. alloca-opt, ... and
AVOIDED_MODULES so the first modules are ordered 1. chmod, 2. btowc,
3. access produces the following diff:
diff --git a/lib/gnulib.mk.in b/lib/gnulib.mk.in
index 711ddcf1260..f16c018b728 100644
--- a/lib/gnulib.mk.in
+++ b/lib/gnulib.mk.in
@@ -34,9 +34,9 @@
# --no-libtool \
# --macro-prefix=gl \
# --no-vc-files \
-# --avoid=access \
-# --avoid=btowc \
# --avoid=chmod \
+# --avoid=btowc \
+# --avoid=access \
# --avoid=close \
# --avoid=crypto/af_alg \
# --avoid=dup \
So the --avoid modules are emitted in the order they are passed to
gnulib-tool, but the actual modules will be alphabetically sorted.
Therefore, I think the correct code would be:
if len(avoids) > 0:
actioncmd += ''.join([f" \\\n# --avoid={x}" for x in avoids])
if len(modules) > 0:
actioncmd += ''.join([f" \\\n# {x}" for x in sorted(modules)])
Seems sort of strange but it produces the correct output for that
test.
Collin
