otherwise generated:
| int
| main ()
| {
| return dlopen ();
wasm-ld: warning: function signature mismatch: dlopen
>>> defined as () -> i32 in /tmp/emscripten_temp_6nz4f0ww/conftest_0.o
>>> defined as (i32, i32) -> i32 in /home/builder/emsdk/upstream/emscripten/cache/sysroot/lib/wasm32-emscripten/libc.a(dynlink.o)
em++: error: running limited binaryen optimizations because DWARF info requested (or indirectly required) [-Wlimited-postlink-optimizations] [-Werror]
and erroneously claims that dlopen is not present, and
configure fails.
Change-Id: Ie54db7c2099787a0b63b76b0b6e3e44750c9046e
Signed-off-by: Michael Meeks <michael.meeks@collabora.com>
Not sure why I now suddenly needed this, too, when I recompiled Poco
for WASM (with different options than before).
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I7221947e0cd2264535b6d34e0640b95d7cfc7cf4
We are logging what we are evaluating as JavaScript later in the
function anyway.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I79972ec585dee128bbcca910def1d46e8134cdc5
Newer Emscripten SDKs no longer provide <sys/epoll.h> and
<sys/inotify.h>. The corresponding functionality is not present, I
assume.
Until now we have been using Emscripten SDK 2.0.31 and that is still
the recommended known-to-work version, but I am experimenting with the
latest, 3.1.30.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I26e89dc38554548aee3ce2dbf6ba352917ba6266
LOKit will be initialised in the lok_init_2() call in lokit_main() in
Kit.cpp. This change also puts setting and getting the LOK_OPTIONS
environment variable in the right order.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Iee6f5adcb60bb887083c67666c2d597a15686bf9
Signed-off-by: Tor Lillqvist <tml@collabora.com>
You now must pass --with-wasm-additional-files=<path> where path
contains a file called sample.docx.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I21e62feb6282833a5d60b31db26328eda63cdaea
Now I hope things are initialised in the right order and the plumbing
gets set up so that messages are passed as expected. It seems to work
most of the time.
Main changes are:
- The online WASM executable is built using the -s MODULARIZE -s
EXPORT_NAME=createOnlineModule options. This means that the WASM
runtime is not automatically initialized and the main() function
is not automatically started. Only when the createOnlineModule()
function is called is that done. Calling exported C/C++ functions
is a little bit more complicated.
- Code to actually Base64-encode strings to be executed as
JavaScript when expected is now present in wasmapp.cpp. (After
being passed through the Base64ToArrayBuffer function on the JS
side.) Whether this is actually necessary is not fully clear, but
to keep the code similar to that in the GTK, iOS, and Android
apps, this is kept as such for now. It would probably work fine to
just directly create the ArrayBuffer in the C++ (using the EM_ASM
magic).
- The COOLWSD::run() function is now run in a separate thread so
that main() can return.
- The FakeWebSocket's onopen() function is now called from
innerMain(), where the HULLO message is sent. It remains a bit
unclear if this really is the ideal place.
In the mobile apps the HULLO message is sent and the onopen()
function is called in the window.socket.onopen() function in
global.js.
But note that despite that the WASM app and the mobile apps are
largely quite similarly constructed and the FakeSocket and
FakeWebSocket plumbing is the same, there is an important
difference. In a mobile app the C++ code is what runs first, and
that then loads the HTML page into WebKit, in which the JS
runs. In the WASM app it is the other way around. The web page is
naturaly the one that is loaded and the JS code then starts
running the C++ code as WASM.
Finally, note that the whole concept that there is a separate "WASM
app" is temporary.
What we eventually want to achieve is that the COOL webpage upon
loading will connect a COOL server. As it does currently. The COOL
server runs the online and core C++ code to load a document, and
renders document tiles and sends those to the client JS code to
dispay.
The new thing will be that, if enabled, in addition to the HTML and JS
resources, the client will also download the WASM code and data
resources. Also, the document and updates to it will be downloaded
while being edited so that a copy can be kept in client memory. But
the WASM code and the downloaded document will remain unused most of
the time. Only if the connection to the COOL server breaks will the JS
start running the WASM code and the JS will talk to online code
running locally as WASM instead of to a COOL server. Obviously there
are still lots of things hanging in the air here regarding how exactly
this will work.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Ib1786a0b485d51797b0f2302d4296aa1ff9df5c1
After my shortly upcoming commits the document gets loaded and its
tiles displayed by the code that normally does that. No need to
separately verify that loading a document works.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Id5883f36644b5e9b79c3f80ea2131e9f35823b2f
Apparently handle_cool_message() gets called from a web worker thread
and then using emscripten_run_script() in wasmapp.cpp to run a JS
snippet that refers to the 'window' variable will not work. That
variable exists only in the main thread. So use MAIN_THREAD_EM_ASM()
instead.
Hardcode the document URL for now also in wasmapp.cpp.
Try to send the HULLO message from COOLWSD::innerMain().
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Ic48042c6d0c6239a3b82e74f0565056a15f3d98d
As we build a single statically linked bunary for WASM, we will end up
with just one copy of each function in it. Both LO core and Poco
include expat. Upstream Poco builds one of the expat source files as
C++. That causes trouble as it then isn't compatible with what LO core
wants to call. Or something like that. You get "RuntimeError: null
function or function signature mismatch". (The "signature" concept in
WASM is orthogonal to C++ name mangling.)
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Iacf75ed34eea98611cb6aa6867e460697ea1fc4a
This revealed some interestring problems related to expat, static
linking, and how expat was built in Poco.
Don't bother saving the document, though.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I6715beb46ad6c4854ffc8aff6a26419c05727ae7
No need to start a separate thread to run it, like in the gtk+
"mobile" app, I think.
Also no need for a cleanup function, I think.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I23e94200047a195b7fecbdf370142daebc3df55c
We want the code to go further and not stop right away.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: Id285f119dde0894cda59b68ebd02a9f6b0e2fbee
Just 4 is very likely not nearly enough. I think I did see some error
that could be related to running out of threads. Does
PTHREAD_POOL_SIZE limit how many threads can exist simultaneously?
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I5da637b5660e89655f7049b6754f70c74dff6da2
Also using soffice.data name instead online.data: online.js
looking for soffice.data
Signed-off-by: Balázs Varga (allotropia) <balazs.varga.extern@allotropia.de>
Change-Id: I62cee3f4866a2824a08b472f15bcdec06a6407b9
* add dependency on soffice.html.linkdeps to rebuild if core was rebuilt
* copy needed data files as-is from core build
* rename executable by setting automake EXEEXT var - appears to work
Signed-off-by: Michael Stahl <michael.stahl@allotropia.de>
Change-Id: I458b49290dae9d621a8043b1b3103d8b8fd606b8
The LO build directory in question needs to be one separately
configured for this (not one that would use Qt5 for UI, for instance),
so in the long run perhaps it does not make sense to create the FS
image there in core, but we should do it here in online? And we will
surely need additional files in the fs image anyway that core knows
nothing about.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I2cff0421da6075eefc017603ddf9d2ecf9dc79e2
It is called from JavaScript with one parameter, the string containing
the message to be sent to the Online C++ code. (See
browser/html/cool.html.m4.) It is not called from the mobile app glue
code.
Signed-off-by: Tor Lillqvist <tml@collabora.com>
Change-Id: I893dfff3b7d2dbfe4bd48393e58a8c6bfc6e2931
Include emsripten headers and init some variables for later.
Signed-off-by: Balázs Varga (allotropia) <balazs.varga.extern@allotropia.de>
Change-Id: Ic2c8228d44a60b25ae495a2e9d75c10160161f11
Also feature/wasm rebased with online master.
Signed-off-by: Balázs Varga (allotropia) <balazs.varga.extern@allotropia.de>
Change-Id: I1ecba4091c22878aacc3d6033ad350b0aa2276dc
Remove the dumb hard-coded home dir and add a section how to use the
pre-built dependencies from the container.
Signed-off-by: Michael Stahl <michael.stahl@allotropia.de>
Change-Id: Ibdc8d766be54dcde440682d2c77fb47f042f6056
-pthreads is required, or wasm-ld reports errors about
"was not compiled with 'atomics' or 'bulk-memory' features"
Also, POCO needs to be built with this, add it in README.
Signed-off-by: Michael Stahl <michael.stahl@allotropia.de>
Change-Id: Ie83e3942e5fc689e6df5a5a705d7ee2e1325ce03
Use the newly introduced soffice.html.linkdeps from core to get the
recursive dependencies into the link command.
This currently fails due to some problem with POCO:
wasm-ld: error: --shared-memory is disallowed by AtomicCounter.o because it was not compiled with 'atomics' or 'bulk-memory' features.
Signed-off-by: Michael Stahl <michael.stahl@allotropia.de>
Change-Id: I76b0a2265f67e89f6992d556525f1263ad1b45db
Copy the list of .cpp files from the Android project, assuming this will
be similar in scope.
Signed-off-by: Michael Stahl <michael.stahl@allotropia.de>
Change-Id: I57c7ad2f10d1867307ff4fcea3d0c650726d18d8
