WinTLSSession buffers received decrypted data into its own buffer. If
read is requested, it copies the data from its buffer. But if
requested buffer size is less than decrypted buffer, some of the data
is left in the buffer. Previously, we had no facility to check the
existence of this pending data. If this data is the last requested
data from remote server, we may end up waiting for read event even if
we have already data in our buffer, which may cause hang. This commit
fixes this issue by introducing function to return the buffered length
in TLSSession. SocketCore also provides the same function, which
delegates to TLSSession object.
This commits implements basic downloading feature for sftp URI. This
includes segmented downloads, and connection (sftp session) pooling.
Download via HTTP proxy has not been implemented yet.
In my quest to explore the code and understand it, why not clean it up
in the progress. Most formatting provided by clang-format. ;)
Aside from formatting, also extracted some method and unnested some
control structs.
Now dedicated ConnectCommand handles connection establishment. It
checks whether connection is established or not. It also handles
backup connection. The next Command creation is abstracted using
ControlChain struct template.
Issue A record query only when non-loopback IPv4 address is
configured. Likewise, issue AAA record query only when non-loopback
and non-linklocak IPv6 address is configured.
The old implementation starts to find faster host when the number of
missing segment becomes 1. Because of --min-split-size option, before
the number of missing segment becomes 1, the number of connection
becomes 1 and it can be slow. In this case, we have to wait until the
last segment is reached. The new implementation starts to find faster
host when the remaining length is less than --min-split-size * 2, to
mitigate the problem stated above.
The old implementation calculates download/upload statistics for a
RequestGroup by summing up all PeerStat objects. For global
statistics, those are summed together. This clearly incurs runtime
penalty and we introduced some kind of caching which updates
statistics every 250ms but it did not work right.
This change removes all these aggregation code, and instead makes
RequestGroup and RequestGroupMan objects hold NetStat object and
download/upload bytes are directly calculated by thier own NetStat.
This is far more simplar than the old way and less runtime penalty and
brings more accuracy.
In CreateRequestCommand, if Request object returned from getRequest()
is still sleeping, CreateRequestCommand pools it back but still holds
its reference. This makes assertion error in
UnknownLengthPieceStroage::hasMissingUnusedPiece() from
AbstractCommand::execute().
The previous implementation constructs proxy URI in OptionHandler but
it cannot handle with the situation when user, password and proxy URI
is given various order. Now we just set rules: username set in
--*-proxy-user overrides username in --*-proxy option, no matter in
any order username, password and proxy URI are parsed. Likewise,
password set in --*--proxy-passwd overrides password in --*-proxy
option.
Tatsuhiro Tsujikawa