This commit adds a field to Batcher to configure the maximum allowed
line length. If this value is non-zero, the bufio.Scanner buffer
will be configured scan lines up to this length. If a line exceeds
this length an ErrLineToLong error will be returned.
Simplified TestBatcher_read tests and added a test case for this
new scenario and to ensure lines exceeding the default
bufio.MaxScanTokenSize are allowed.
I did this with a dumb editor macro, so some comments changed too.
Also rename root package from platform to influxdb.
In interest of minimizing risk, anyone importing the root package has
now aliased it to "platform" so that no changes beyond imports were
necessary in those files.
Lastly, replace the old platform module to local path /dev/null so that
nobody can accidentally reintroduce a platform dependency while
migrating platform code to influxdb.
In the case that there is a read error, we would close the lines
channel before sending the error into the read error channel. closing
lines then allows the write goroutine to possibly send in a nil error
before read is able to, causing the main function driving both to
return a nil error. Additionally, it is possible for both reads and
writes to race sending errors into their channels, and the main
goroutine will only read from one, causing the other goroutine to leak.
To fix this, we close lines only after we have sent an error into
the channel, we ensure we read from both errors to make sure that
both have exited, and we unify the channels and add a buffer of size
two to the channel. It is possible for write to exit leaving read blocked
forever, but write only exits with a nil error when read has exited, so
this only happens during an actual write error, just like before.
Channels are hard.
Pavel Zavora
Greg
Stuart Carnie
Jonathan A. Sternberg
David McKay
Mark Rushakoff
Jeff Wendling
Chris Goller