aboutsummaryrefslogtreecommitdiff
path: root/Makefile
diff options
context:
space:
mode:
authorDavid Baker <dave@matrix.org>2018-09-25 17:13:29 +0100
committerDavid Baker <dave@matrix.org>2018-09-25 17:13:29 +0100
commit263b94428a24caaa5b899ed7f73b896620e6cdf4 (patch)
treedacfa7322ee472d9a09e4bd56805bae0aad215c9 /Makefile
parentf29d8cdd7bf1faf8294f51624c633fb05c9a0e2f (diff)
Another day, another interface
Change the interface again, hopefully this time a bit more normal. Now we wrap the emscripten module completely and just expose the high level objects. The olm library export is now imported as normal (ie. returns a module rather than a function returning a module) but has an `init` method which *must* be called. This returns a promise which resolves when the module is ready. It also rejects if the module failed to set up, unlike before (and unlike the promise-not-a-promise that emscripten returns). Generally catch failures to init the module.
Diffstat (limited to 'Makefile')
-rw-r--r--Makefile28
1 files changed, 27 insertions, 1 deletions
diff --git a/Makefile b/Makefile
index dcd5cc1..d99c8fc 100644
--- a/Makefile
+++ b/Makefile
@@ -41,11 +41,22 @@ FUZZER_BINARIES := $(addprefix $(BUILD_DIR)/,$(basename $(FUZZER_SOURCES)))
FUZZER_DEBUG_BINARIES := $(patsubst $(BUILD_DIR)/fuzzers/fuzz_%,$(BUILD_DIR)/fuzzers/debug_%,$(FUZZER_BINARIES))
TEST_BINARIES := $(patsubst tests/%,$(BUILD_DIR)/tests/%,$(basename $(TEST_SOURCES)))
JS_OBJECTS := $(addprefix $(BUILD_DIR)/javascript/,$(OBJECTS))
+
+# pre & post are the js-pre/js-post options to emcc.
+# They are injected inside the modularised code and
+# processed by the optimiser.
JS_PRE := $(wildcard javascript/*pre.js)
JS_POST := javascript/olm_outbound_group_session.js \
javascript/olm_inbound_group_session.js \
javascript/olm_pk.js \
javascript/olm_post.js
+
+# The prefix & suffix are just added onto the start & end
+# of what comes out emcc, so are outside of the modularised
+# code and not seen by the opimiser.
+JS_PREFIX := javascript/olm_prefix.js
+JS_SUFFIX := javascript/olm_suffix.js
+
DOCS := tracing/README.html \
docs/megolm.html \
docs/olm.html \
@@ -67,6 +78,15 @@ EMCCFLAGS = --closure 1 --memory-init-file 0 -s NO_FILESYSTEM=1 -s INVOKE_RUN=0
# longer needed.
EMCCFLAGS += -s NO_BROWSER=1
+# Olm generally doesn't need a lot of memory to encrypt / decrypt its usual
+# payloads (ie. Matrix messages), but we do need about 128K of heap to encrypt
+# a 64K event (enough to store the ciphertext and the plaintext, bearing in
+# mind that the plaintext can only be 48K because base64). We also have about
+# 36K of statics. So let's have 256K of memory.
+# (This can't be changed by the app with wasm since it's baked into the wasm).
+EMCCFLAGS += -s TOTAL_STACK=65536 -s TOTAL_MEMORY=262144
+
+
EMCC.c = $(EMCC) $(CFLAGS) $(CPPFLAGS) -c
EMCC.cc = $(EMCC) $(CXXFLAGS) $(CPPFLAGS) -c
EMCC_LINK = $(EMCC) $(LDFLAGS) $(EMCCFLAGS)
@@ -147,13 +167,19 @@ $(STATIC_RELEASE_TARGET): $(RELEASE_OBJECTS)
js: $(JS_TARGET)
.PHONY: js
-$(JS_TARGET): $(JS_OBJECTS) $(JS_PRE) $(JS_POST) $(JS_EXPORTED_FUNCTIONS)
+# Note that the output file we give to emcc determines the name of the
+# wasm file baked into the js, hence messing around outputting to olm.js
+# and then renaming it.
+$(JS_TARGET): $(JS_OBJECTS) $(JS_PRE) $(JS_POST) $(JS_EXPORTED_FUNCTIONS) $(JS_PREFIX) $(JS_SUFFIX)
EMCC_CLOSURE_ARGS="--externs $(JS_EXTERNS)" $(EMCC_LINK) \
$(foreach f,$(JS_PRE),--pre-js $(f)) \
$(foreach f,$(JS_POST),--post-js $(f)) \
-s "EXPORTED_FUNCTIONS=@$(JS_EXPORTED_FUNCTIONS)" \
-s "EXTRA_EXPORTED_RUNTIME_METHODS=$(JS_EXTRA_EXPORTED_RUNTIME_METHODS)" \
$(JS_OBJECTS) -o $@
+ mv $@ javascript/olmtmp.js
+ cat $(JS_PREFIX) javascript/olmtmp.js $(JS_SUFFIX) > $@
+ rm javascript/olmtmp.js
build_tests: $(TEST_BINARIES)