GitLab

@jbenet @whyrusleeping

this addresses a failure case where

1) bitswap wants blocks A and B
2) partner 1 has A and partner 2 has B
3) We choose a key at random, drawing A.
4) Then, we request A, neglecting to find a provider for B.

Sending the full wantlist is meant to be used as a helpful additional
piece of data, but...

unless our hunch is support by statistical inference at runtime,
it's not safe to assume that a peer will have blocks for related keys.
Routing must be the source of truth.

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

oops

set Add is idempotent but it's a waste of resources

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@jbenet @whyrusleeping

the pyramids were built one brick at a time

addresses: https://github.com/jbenet/go-ipfs/issues/370

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping @jbenet

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping i hope this makes it a bit easier to work with tests

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping @jbenet

since the logger is created with package scope, don't need to specify
the package name in event messages

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@jbenet
@whyrusleeping

Let me know if you want to direct the eventlog output to _both_ the file
and stderr. Right now it goes to file. Perhaps this is just a minor bip
in the larger discussion around log levels.

https://github.com/jbenet/go-ipfs/issues/292

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@whyrusleeping This appears to be a timing issue. The asynchronous
nature of the new structure provides has the bitswap waiting on the
context a bit more. This isn't a problem at all, but in this test, it
makes the functions return in an inconveniently timely manner.

TODO don't let the test depend on time.

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

vanilla:
    21.57 real        45.14 user         8.51 sys

short:
    14.40 real        31.13 user         5.56 sys

License: MIT
Signed-off-by: Brian Tiger Chow <brian@perfmode.com>

@jbenet

remove 'adapter' concept
instead, describe the component as the bitswap network

it's still an adapter, but it's just not necessary to describe it as
such