git.iko.soy/distributed-process
on 87352cc help sign in
87352ccminor: allow args passed to install_allRyan Newton 29bedf5Added warning message to the tryUntilNoIOErr loop.Ryan Newton 699ff8fMerge branch 'master' of github.com:haskell-distributed/distributed-processRyan Newton 4e42a1eMerge pull request #6 from acfoltzer/masterRyan Newton 48e257aCleaned up examples and benchmarks. Found strange bug. Tweaked Process.hs for strict bytestrings.Ryan Newton 4341d5dfixed up demotransportAdam Foltzer 5395dd9Merge branch 'master' of https://github.com/haskell-distributed/distributed-processAdam Foltzer 5994848added pragma for distributed-process to handle lazy/strictAdam Foltzer ff8d9c6Add SendTransport related entries in benchmarks/Makefile and performed some refactoringRyan Newton 97352daAdd a simple install_all script for convenienceRyan Newton 5e0b8b2Merge branch 'master' of github.com:haskell-distributed/distributed-processRyan Newton cdfd7daCleaned up the benchmark setup. Now all benchmarks can run either a single trial (no criterion) or multiple trials with criterion.Ryan Newton acbc141Merge pull request #5 from acfoltzer/masterNicolas Wu f5494b3strictified network-transport-pipesAdam Foltzer 900dea9merged upstream changes with lazy/strict switch and bug fixesAdam Foltzer fc72a01tweaked docs and Send benchmarkAdam Foltzer 595ed74merged changes from homeAdam Foltzer bb31a08added switch between lazy and strictAdam Foltzer 30147dbFixed a major bug in recvExact; the number of bytes we request from recv has to be bounded. Right now it will just ask for the min of the remaining message size and 4096.Adam C. Foltzer 8bdb68cswitched back to Char8 bytestrings; enhanced sendtransport so it's not just sending all 0sAdam C. Foltzer 9ba9186fixed bugs in PingTransport benchmark and TCPAdam C. Foltzer 931436beta-expanded error helpersAdam C. Foltzer 32e3b04use vectored IO rather than mapM to send size header with messageAdam C. Foltzer 1e377abefficient implementation of recvExact using bytestring internalsAdam C. Foltzer 2ce3419strictified, switched back to cereal in preparation for using conduitsAdam C. Foltzer 95bf372Factored pipes backend into its own package. One reason for this is that it introduces some additional dependencies, and we always want to minimize dependencies.Ryan Newton e7d5c21Got the Pipes backend working with unix-bytestring and normal unix.Ryan Newton 506d4baFixed the pipes transport so it opens the file only once on the reader side. I'm observing a 72 us latency for PingTCP (vs. 28us for the equivalent C program). I'm observing a 99 us latency for PingPipes, and it goes up to 100% CPU. That's with the *String* based Pipes, however!Ryan Newton 6995d1fTweak makefile and add another for benchmarksRyan Newton fa27f80Minor, add warning at the top of Pipes about it being unfinishedRyan Newton af7778cAdded a close routine for Pipes that removes the named pipe from diskRyan Newton 30fc293Cleaned out all unneed comments and debugging printfs in Pipes.hsRyan Newton 06812deClean up demos file and make small update to READMERyan Newton 89b2116Fixed remaining demo. Pipes works for all 5.Ryan Newton c2f03d8Made Pipes use nonblocking mode and changed a number of things to make this happen. Now all three transports pass all demos except demo2.Ryan Newton d72c0f5Got to the bottom of the regression in the Pipes transport. The catalyst was switching between cereal and binary. I have no idea (still) why this is.Ryan Newton 8c065e4minor: make writes use strict bytestrings to reduce interleavingRyan Newton cc13eecAdd more exception handling. Pipes is still failing outright, the listener reads the header of the first message, then attempts to read the payload and gets stuck: 'too few bytes'.Ryan Newton 913daccTwo small changes to TCP transport. Added propagation of exceptions from child threads. Added more defensive pattern matching for the one byte read.Ryan Newton 62eb0b1Added a NOOP closeTransport to the MVar transport. It passes all demos tooRyan Newton 0c0f406After merge making sure that TCP transport works for all demos. All demos enabled in this rev, TCP transport only.Ryan Newton ad9c9ccThe Pipes transport works for demo1 and demo0.Ryan Newton 7041cc4Merge branch 'master' of git://github.com/haskell-distributed/distributed-processRyan Newton 6c768b5Begin to update DemoTransportRyan Newton 7cc886fFactored Pipes.hs to work with either unix or unix-bytestring. Bytestring versions should be more efficientRyan Newton f3b2812Removed some of the unused options now that something is working. Both opening up the writing end immediately and the speculative strategy (opening it on another thread) work.Ryan Newton 1611171This is still covered in debugging cruft, but it works without opening and closing the named pipe on every send.Ryan Newton 12559abThis checkin passes a simple demo test, but it still depends on opening and closing the fifo on every send. This is proving trickier to get right than I thought.Ryan Newton 8c57693Added a simple named Pipes transport. Currently in debugging mode.Ryan Newton ff10a6eImprove recvExact and add SCCs.Nicolas Wu