Now you can't use the same syntax as you would for a method to
implement a property as whole, instead you need to specify the
getter and/or setter explicitly. This is to allow parent classes
to expand their properties without altering behavior of the child
classes.
so here's the ugly problem. libproxy. yes. we've discussed memory
usage (e.g. it may have to execute javascript and pull in lots of deps
etc.) but we dlopene'd on the fly. ok... but this didn't solve another
issue i hit:
libproxy was causing enlightenment to abort(). some internal bit of
libproxy was raising a c++ exception. this wasn't caught. this causes
an abort(). takes down your entire desktop. FANTASTIC. this is bad. i
wouldnt' expect a library we depend on to be THIS anti-social but
libproxy seemingly is. it SHOULd catch its error sand just propagate
back to us so we can handle gracefully.
there reall is no way around this - isolate libproxy. it's even worse
that libproxy can load arbitrary modules that come from anywhere sho
who knows what issues this can cause. isolation is the best solution i
can think of.
so this makes an elf+net_proxy_helper we spawn the first time we need
a proxy lookup. we re-use that binary again and again until it exits
(it should exit after 10 seconds of being idle with no requests coming
in/pending). it'll respawn again later if needed. this involves now
the efl net threads having to marshall back to mainloop to do the
spawn and to write to the proxy process (reading is done by async exe
data events and the data is passed down a thread queue to the waitng
efl net thread). if the exe dies with pending requests unanswered then
it's respawned again and the req's are re-sent to it... just in case.
it has a limit on how often it'll respawn quickly.
this seems to work in my limited testing. this ALSO now isolates
memory usage of libproxy to another slave process AND this process
will die taking its memory with it once it's been idle for long
enough. that;s also another good solution to keeping libproxy impact
at bay.
this fixes building against openssl 1.1 since it broke api in various
ways by hiding structs and deprecating api's (this causes warnings not
breaks unlike the struct hiding). this adapts to these changes and
makes efl build again.
@fix
The _c structure used to store the Curl function pointers needs
referencing to be sure it is not freed although still needed.
The non-referencing was leading to a crash during the destruction
of the dialers. The _c was still used although it has been freed
by ecore_con_url_shutdown.
While a socket can be closed to receive data resulting in EOS, it
could still be used to send stuff. Then it won't result in "finished",
just "read,finished" event.
However, previously this was considered a disconnect and we must
respect this otherwise tests (Ecore_Con_Eet suite) will hang waiting
for a disconnect.
On destructor we're not supposed to emit events, I even thought that
would be implicit, but it's not. If we do, for example an event
handler that would 'efl_del()' on "EFL_IO_CLOSER_EVENT_CLOSED" would
trigger too-many unrefs.
UDP client has no socket on its own, so it can't be marked as
close-on-exec, however failing to resolve the method is bad as
confuses the user.
Instead provide a method that checks if the setting is different from
parent, in such case provide a meaningful message. Otherwise just be
quiet.
remove one more TODO: since Efl.Net.Ip.Address was introduced we can
now expose Efl.Net.Socket.Udp.init as a protected method that will
configure the internal address we use for the remote peer. This allow
subclasses to override or call such methods.
There are no jobs to stop there, it was a left over since this code
was somehow derived from Efl.Loop_Fd where "read" and "write" would
constantly be called until the kernel flag was cleared, there the
callback must be removed to stop jobs, not here.
Working directly with Eina_Slice is easier than a pointer to it,
requires no validation of the pointers and is cheap since it's just
putting together size_t + void*.
However we can't hint the user of 'const(Eina.Slice)' properties as
Eolian is incorrectly generating getters as:
const Eina_Slice class_property_get(...)
which is makes compilers complain about ignored qualifiers:
../src/lib/ecore/efl_io_copier.eo.h:329:7: warning: 'const' type qualifier on return type has no effect [-Wignored-qualifiers]
Leave some TODO so @q66 can fix those.
Previously we couldn't return a slice, instead required the user to
pass a slice and we'd fill it since Eolian couldn't generate fallbacks
for structures.
Since @q66 fixed eolian, we can now return the structure itself as
initially wanted, ditching some TODO from the code.
Some applications will create the handle, immediately send data, flush
and delete it, expecting the data to be sent to remote peer.
This is a bad behavior as the application would become unresponsive
until the connection is established, data can be written (since
depends on server consuming it), then allow it to be closed.
A proper behavior here would be to chain based on events, with the
usage of a copier would be simply wait for "done" event.
However the legacy API allowed this and terminology depends on this
awkward "feature", thus be bug-compatible.
This fixes T5015.
In the old/legacy API the socket would be opened early in non-blocking
mode (connect returned errno==EINPROGRESS), with UNIX socket being
path-validated early and returning NULL as 'server' handle.
Some applications relied on this instead of monitoring the "ERROR"
events, considering the connection to be successful if there was a
handle -- this was the case with Terminology after it moved from DBus
to Ecore_Ipc.
Although this is not correct, we must keep compatibility and thus we
stat() in compatibility layer, failing early as the old API would do.
These legacy API had the nasty behavior of keeping handles alive until
the pending events were dispatched, this could happen after the module
itself was shutdown, resulting in log to unregistered domains.
Then do not unregister the domain -- eina_shutdown will avoid leaks
anyway.
We do not need to keep a "only_head" flag, but we must set
CURLOPT_NOBODY instead of going the "CUSTOMREQUEST" route, otherwise
curl won't follow redirects, etc.
With the last patch to fix delete-from-curl callback it went too much,
considering it was always dead (in the test scenario it was, so it was
"right"), but broke normal cases.
This is a string parser, serializer and asynchronous resolver.
It's purpose is to convert to and from the strings we use in our
dialers and servers, such as "127.0.0.1:1234" or "[::1]:1234",
properties allow to check the family, port, address bytes (slice) and
even get a struct sockaddr pointer to use with bind()/connect() in
outside code.
It will also offer some utilities present in netinet/in.h in an easy
to use way, after all IN6_IS_ADDR_LOOPBACK() works one way, while
there is no IN_LOOPBACK and comparing with INADDR_LOOPBACK will lead
to errors since it's in network order.
Last but not least, it will do asynchronous resolve of host and port
names using an internal thread and getaddrinfo(). The results are
delivered using a Future with an array of objects.
This is a major work and unfortunately couldn't be split into smaller
pieces as old code was highly coupled.
Ecore_Con_Server is now a wrapper around Efl_Net_Dialer_Simple
(ecore_con_server_connect()) and Efl_Net_Server_Simple
(ecore_con_server_add()), doing all that the original version did with
some fixes so ecore_con_ssl_server_upgrade() and
ecore_con_ssl_client_upgrade() are more usable -- see the examples and
-t/--type=tcp+ssl.
I tried to be bug-compatible, with code annotations where things
doesn't make sense. This was based on ecore_con_suite tests and some
manual experimenting with the examples, these can be helpful if you
find regressions (report/assign to me).
if we're doing 30x redirects or anything else, keep going. If it's an
error it will be reported later, otherwise EOS. Fact is we're only
interested if 20x.
This was annoying to identify as the sequence is kinda difficult to
get, but Terminology was doing a HEAD request and it was triggering
this case in particular.
Fixes T4975.
Windows time_t is not a long, but long-long, then stick with int64_t
so it works everywhere (converts to time_t internally).
And there is no gmtime_r(), then use the gmtime() if not detected.
If we can parse the IP using inet_pton() and the port, there is no
reason to call getaddrinfo() in a thread.
This is required since ecore_con_suite (for ecore_con-over-efl_net) will
assume the server is running as soon as it's created.
If we want to upgrade a dialer, then we must have a way to know if
that socket has already adopted another socket so we don't create it.
We can't simply use efl_net_socket_ssl, otherwise we'd miss some
methods such as efl_net_dialer_address_dial_get() and events such as
connected.
if we create an object, say a TCP dialer, and don't connect/bind, then
we have no FD (=0). If we set FD to INVALID_SOCKET on start, other
parts of the code will fail since they consider that "closed", but
we're not closed yet.
Then check for family == AF_UNSPEC && fd == 0, if so don't close it.
At least in ArchLinux the function has no "_" in the symbol name,
matching perfectly what's in the header.
If in other systems it misses such symbol, then check for both.
Rewrite Ecore_Con_Url as a non-Eo (since it's just legacy) that is
built on top of Efl_Net_Dialer_Http.
Since there are some legacy behavior we do not want to expose in the
new classes, hack around and manipulate the curl_easy_setopt()
directly in those cases.
This includes the cookies: there is no reason why we should expose
independent files for read (COOKIEFILE) and write (COOKIEJAR), real
world applications can manipulate the files directly, like copying
from a template to a RDWR before using, etc.
CURL needs some special curl_easy_setopt() calls to enable automatic
gzip deflate (CURLOPT_ENCODING) and
If-Modified-Since/If-Unmodified-Since logic.
As If-Modified-Since/If-Unmodified-Since requires a timestamp string,
let's expose class methods to handle those.
Summary:
Currently eolian abbreviates when only the last word of class name and
the first word of method name are same, but this patch abbreviates
generated c name of function to remove all duplicated affix.
For example, "efl_io_closer_fd_closer_fd_set" will be "efl_io_closer_fd_set".
Reviewers: jpeg
Subscribers: cedric
Differential Revision: https://phab.enlightenment.org/D4430
Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
The low level I/O primitives are powerful but adds some complexity to
use, for bi-directional streaming communication one ends creating two
Efl.Io.Queue and two Efl.Io.Copier to pipe data to socket when it can
operate.
Then encapsulate the socket using the new Efl.Io.Buffered_Stream, this
will allow the socket, be a dialer or a server client, to be operated
as a single handle that internally carries about the buffering for
you.
As one can see in the examples, compared to their "manual"
alternatives they are very easy to use, ressembling
Ecore_Con_Server/Ecore_Con_Client, but also offers line-based
delimiters and the possibility to let the socket to handle queueing
for you in case you received partial messages (just do not
read/clear/discard the received data).
I just realized that if a client is not referenced it would leak in
the 'ssl' server as we must del it.
However, if we del the SSL socket, we're going to close the underlying
TCP. But we're from the TCP "client,add" callback and this causes
issues since "closed" will be emitted, our close callback will
unparent the client, which lead to it being deleted.
The proper solution is to only monitor "closed" if the client is
accepted. Otherwise we just check if it was closed, if we're the
parent, etc...
Fixing this in all servers were painful, we could share since most
inherit from Efl.Net.Server.Fd. Then add the "client_announce"
protected method to do it, and document how it should work.
There was a bug that if the remote peer closed the connection, it
would trigger 'read' event, which would read 0 bytes, flagging as
EOS... but then marking as "can_read", which was wrong.
Just stop monitoring the events and fix that.
If we let the user know he can read or write, stop monitoring
otherwise fd handler will constanly report of data to read/write until
its actually done, which would clear the kernel flag.
Since we use "can_read" and "can_write" for that, toggle the callback
connection that manages the actual Ecore_Fd_Handler monitor.
Instead of adding a job to create the socket and call bind(), do it
straight from the serve() method, this allows the caller to set
umask(), permissions and so on.
Document this behavior in the class, since we can't extend the
method's documentation.
The new efl_net code won't compose any path own its own, allowing the
user to connect to non-EFL systems.
However we need a way to use the same path Ecore_Con_Server does, so
we can reach it. Then export and use ecore_con_local_path_new() to do
exactly that.
Summary:
fix warnings while generating documents
- end of file while inside a group (eina_util.h)
- missing title after \defgroup
- ignoring title "Ecore_Con_Lib_Group" that does not match old title
Reviewers: Hermet
Subscribers: cedric, jpeg
Differential Revision: https://phab.enlightenment.org/D4420
Since connman is specific to linux, on other platforms just compile a
dummy "none" backend that will always report online and no other
details. This will be used in Windows, MacOS and other platforms that
still lack a proper backend.
The compile-time infrastructure also allows for networkmanager to be
added with ease, simply copy "efl_net*-none.c" or "efl_net*-connman.c"
to be a starting point and then add its specifics, adapting
configure.ac and Makefile_Ecore_Con.am
The time to live hop limit should not be named loopback and have a type that
can actuall hold the number of hops. It already was always uint8 in the code.
Just fix the eo file.
windows is nasty and defines the value to be set or retrieved as
'char *', which triggers a warning when we use another kind of
pointer.
Partially addresses D4357.
windows is nasty and defines the payload to be sent or received as
'char *', which triggers a warning when we use another kind of
pointer.
Partially addresses D4357.
On Windows SOCKET is unsigned, thus will cause sign errors when
formatting with "%d" or comparing with signed values.
On UNIX it was quiet and easy to miss, thus a new #define can be used
to check for those. It will use 'unsigned long' as SOCKET, thus will
complain out loud and not even work correctly when using pointers on
64bits UNIX on mistakes -- which should improve the situation.
This helped to fix lots of missing conversions, all fixed.
This partially addresses D4357.
printf() was not happy with a void* being used with the %s format.
Since the 'key' values are actually strings and let's declare them
as const char* to fully embrace the string semantics.
These are objects to allow control of networking devices
(efl_net_control) as well as an application to request for
connectivity (efl_net_session).
They are loosely based on ConnMan.org, which we already use in
Enlightenment Window Manager via DBus access with Eldbus. However they
do not map 1:1 as the goal was to expose a viable subset of controls
but in a simple and general way, thus nome strings were converted to
enums, some arrays of strings were converted to bitwise flags, some
names were made more general, such as "service" was turned into
"access point" so it doesn't generate confusion with other "network
services" (ie: http server), or "favorite" that was renamed to
"remembered". Some behavior are slightly different (yet able to be
implemented on top), such as "Service.MoveBefore" and "MoveAfter" were
converted to a numeric "priority", calculated from service's list
index, changing the priority will reoder the list and thus generate
the MoveBefore and MoveAfter DBus commands.
ConnMan was chosen not only because we already use it, but because its
DBus API is sane and simple, with the server doing almost all that we
need. This is visible in the efl_net_session, which is completely done
in the server and do not require any extra work on our side -- aside
from talking DBus and converting to Eo, which is a major work :-D
NOTE: ConnMan doesn't use FreeDesktop.Org DBus interfaces such as
Properties and ObjectManager, thus we cannot use
eldbus_model_object.
There are two examples added:
- efl_net_session_example: monitors the connection available for an
application and try to connect. You need a connman compiled with
session_policy_local and a configuration file explained in
https://github.com/aldebaran/connman/blob/master/doc/session-policy-format.txt
to get a connection if nothing is connected. Otherwise it will just
monitor the connectivity state.
- efl_net_control_example: monitors, plays the agent and configure
the network details. It can enable/disable technologies, connect to
access points (services) and configure them. It's quite extensive
as allows testing all of ConnMan's DBus API except P2P (Peers).