In Restbed one is not supposed to call session->yield outside the
threads controlled by Restbed. RetroShare JSON API async call were
calling session->yield from threads controlled by RetroShare all the
times, this caused crashes in some cases, like when the JSON API
socket timed out concurrently with the session->yield call .
To solve this problem session->yield from async
calls are now wrapped insto mService->schedule to ensure they are
executed on the right thread (aka one of the threads controlled by
Restbed).
While solving this issue I realized also that passing RsEvents as const
references around was quite limiting in cases where the event need to
be finally handled in another thread, in that case passing by const
reference the RsEvent needed to be copied by value into the thread
that process it, in this copy by value process the information of
which was the original specific type is lost, and then only the data
and methods from general RsEvents are available, unless the handler
does tricky stuff with type coercion etc. To solve this limitation
pass the events as std::shared_ptr<const RsEvent> seems the safer and
more elegant solution.
JavaScript represents all numbers in a double like manner thus it
treats 64 bit integers in an unsafe manner, to workaroud this problem
provide an alternative strin representation in JSON serialization
format so JavaScript clients can access the correct value someway.
As JSON have no probles with 64 bits integers keep supporting the integer
representation too.
See https://stackoverflow.com/a/34989371
JSON API user would bet an unusable base64 blob due to RsTlv opaque JSON
serialization, this way the content of the set is serialized plainly
so it is usable through JSON API, without breaking binary
serialization compatibility.
We must stop using those RsTlv derived items.
Mails sent in the past end never received were lingering in the outgoing
queue forever as displayed by statistics. If a mail is older then
GXS_STORAGE_PERIOD all nodes delete it from the GXS group so it can
never arrive to destination, there is no point in keeping it in the
outgoing queue, instead notify sending timed out and delete the mail.
Sending PGP fingerprint trough broadcast discovery is misleading in
terms of security, as those packets cannot be trusted for PGP
fingerprint verification, removing it reduce packet size and reduce the
risc of wrong usage of untrusted information
