Now user can set directory to store sessions somewhere else
other than the folder where the program's executable is.
(Based on patch from Felony <onyfel@gmail.com> contributed
in https://github.com/sochix/TLSharp/pull/874 )
People are reporting that under some circumstances, an infinite
loop could happen when TLSharp tries to handle a reconnection to
a different DC, if the DC instructed to be used is the same as
the one that was used in the last connection.
Not sure how could this happen (although the analysis present in
this github issue [1] might help understand it), but this commit
helps to make TLSharp fail fast (with an exception) instead of an
infinite loop from now on, which will help avoiding people file
issues such as [2] and [3] and instead maybe file a proper bug
report easier to understand, to try to fix the underlying root
cause.
[1] https://github.com/sochix/TLSharp/issues/719
[2] https://github.com/sochix/TLSharp/issues/803
[3] https://github.com/sochix/TLSharp/issues/839
It wasn't being used internally, and the use-case to be used
externally is too intricated. There's already a way to change
the handler at contructor time[1] so that should be enough
customization.
[1] 48077961ae
If a method that returns bool is never returning false,
it's essentially not giving any information at all, which
means that it should return void. This will help the
consumers of this API to stop thinking that the result
of this function means if the connection was succesful
(which it is, actually, but it was never returning false
because if there's any problem connecting, the result would
be an exception instead).
The method IsPhoneRegisteredAsync() was calling
RequestWithDcMigration() which was the real method
that made use of the _sender field, so we can move
the null check to there.
This will also help track down possible NullReferenceExceptions
that could happen when other callers of this method
arrive with a bad _sender, which might aid us trying
to fix#706.
Load() method in ISessionStore requires public constructor of class Session in some cases.
E.g., loading session from a db or a key-value store, manual loading from text-file, JSON/XML and so on.
Doing this is better when looking at the problem from at least
these 2 points of view:
a) You're working on TLSharp itself: You might be testing some
new things or running TLSharp's tests. Suddenly, if a FLOOD_WAIT
response happens, there's no clear way to know. You just see the
tests taking forever. But if a test has reached a situation in
which Telegram has returned a FLOOD_WAIT error, it's very likely
that what happens is that the TLSharp operation being tested is
actually buggy, which means that the test should FAIL FAST and
show a stacktrace of the problem so that you can see where in the
code was the FLOOD_WAIT received/caused. You shouldn't need to
kill the run of the test suite and hope to hit the problem again
only when you were using the debugger (to be able to pause
execution and examine a stacktrace of where the execution flow is).
b) You're using TLSharp: if you hit a FLOOD_WAIT problem it may
happen because:
b1) TLSharp has a bug: in this case it's better to throw an
exception so that the user can copy the stacktrace and paste
it into a new Github issue.
b2) Your program uses TLSharp sending excessive requests: you
want to have your program know when you hit the limit, to be
able to fix your program to not be so floody. But a call to
Thread.Sleep() doesn't help you to know this, you just know
that suddenly your program has hung, and you don't know why.
You cannot react to the problem, however with an exception you
can react to the problem (for example by examining the time
that the exception provides, as a TimeSpan property, to know
how much your program needs to wait to be able to use TLSharp
again).
When refactoring recently the PHONE_MIGRATE_X error
wrt exception handling [1] I removed the dubious/obsolete
"settings" part, but I mistakenly removed the URL which
tells library consumers that if they face this exception,
it's actually a bug of the library that they should report.
[1] 77867b44e6
There are different kind of MIGRATE errors that could be
thrown by the Telegram API, as evidenced by this recent
change:
b06f8a8e11
So this rename tries to make it consistent to the new
exception names:
* FILE_MIGRATE_x -> FileMigrationException
* USER_MIGRATE_y -> UserMigrationException
* PHONE_MIGRATE_z -> PhoneMigrationException
