Bug#728584: [Pkg-haskell-maintainers] Bug#728584: cabal-install hangs
Dear Joachim and all mentainers: I have more information on this bug. When I changed ~/.cabal/config file to use remote-repo: old.hackage.haskell.org:http://old.hackage.haskell.org/ instead of remote-repo: hackage.haskell.org:http:// hackage.haskell.org/packages/archive cabal update successed; When I changed back to default, the issue is then reproducable again. In both case, proxychains cabal update will get segmentation fault. Hope it helps. Best, Zengbo 2013/11/4 张增波 zengbo.zh...@gmail.com HI Joachim and all, I guess two situations may help to reproduce it: 1. I am behind two NATs, one is a home router, one is the connection point from ISP to Internet (maybe more, depends on ISP). 2. The network speed is slow from my pc to hackage server (about 10KB/s) And there is another phenomenon after I reported the bug: I tried again (3 times) to do cabal update, with two times get partial content (almost full, let me say, only ~200k lost compared to a wget-ed and verified version but I didn't record the number ) after about 10 minutes then it reported the tarball is broken; and the last time it hangs with 3 hours so I did Ctrl-C to exit (same as the bug report itself). If you need I can do more times and give some accurate data (time/size). From the new phenomenon, I guess again it is about a weak design of tcp itself: when there is only one direction of data flow, the sender may close the tcp link more quickly than the receiver at certain condition: for some point the sender lose all data packages (and re-transmissions) to the receiver, and the receiver don't know it at all. Although it is a tcp weakness, AFAIK, this should be overcome by some upper layer protocol implementation. And another thing I am surprised is about proxychains, it is 100% segmentation fault on my side. I just use a ssh dynamic port forwarding and in wget case it can speed up the download to hackage server. It should be easy to reproduce in my opinion. Best, Zengbo 2013/11/4 Joachim Breitner nome...@debian.org Hi, Am Sonntag, den 03.11.2013, 19:41 +0800 schrieb Zhang Zengbo: I am a new user of haskell-platform. After I installed it, I simply run: cabal update, it hangs, then I try to use --verbose=3 to see what happend: unfortuately, I cannot reproduce it here. Is there anything special about your network setup? E.g. forced proxies or something of the kind? Greetings, Joachim -- Joachim nomeata Breitner Debian Developer nome...@debian.org | ICQ# 74513189 | GPG-Keyid: 4743206C JID: nome...@joachim-breitner.de | http://people.debian.org/~nomeata
Bug#728584: [Pkg-haskell-maintainers] Bug#728584: cabal-install hangs
Hi, Am Sonntag, den 03.11.2013, 19:41 +0800 schrieb Zhang Zengbo: I am a new user of haskell-platform. After I installed it, I simply run: cabal update, it hangs, then I try to use --verbose=3 to see what happend: unfortuately, I cannot reproduce it here. Is there anything special about your network setup? E.g. forced proxies or something of the kind? Greetings, Joachim -- Joachim nomeata Breitner Debian Developer nome...@debian.org | ICQ# 74513189 | GPG-Keyid: 4743206C JID: nome...@joachim-breitner.de | http://people.debian.org/~nomeata signature.asc Description: This is a digitally signed message part
Bug#728584: [Pkg-haskell-maintainers] Bug#728584: cabal-install hangs
HI Joachim and all, I guess two situations may help to reproduce it: 1. I am behind two NATs, one is a home router, one is the connection point from ISP to Internet (maybe more, depends on ISP). 2. The network speed is slow from my pc to hackage server (about 10KB/s) And there is another phenomenon after I reported the bug: I tried again (3 times) to do cabal update, with two times get partial content (almost full, let me say, only ~200k lost compared to a wget-ed and verified version but I didn't record the number ) after about 10 minutes then it reported the tarball is broken; and the last time it hangs with 3 hours so I did Ctrl-C to exit (same as the bug report itself). If you need I can do more times and give some accurate data (time/size). From the new phenomenon, I guess again it is about a weak design of tcp itself: when there is only one direction of data flow, the sender may close the tcp link more quickly than the receiver at certain condition: for some point the sender lose all data packages (and re-transmissions) to the receiver, and the receiver don't know it at all. Although it is a tcp weakness, AFAIK, this should be overcome by some upper layer protocol implementation. And another thing I am surprised is about proxychains, it is 100% segmentation fault on my side. I just use a ssh dynamic port forwarding and in wget case it can speed up the download to hackage server. It should be easy to reproduce in my opinion. Best, Zengbo 2013/11/4 Joachim Breitner nome...@debian.org Hi, Am Sonntag, den 03.11.2013, 19:41 +0800 schrieb Zhang Zengbo: I am a new user of haskell-platform. After I installed it, I simply run: cabal update, it hangs, then I try to use --verbose=3 to see what happend: unfortuately, I cannot reproduce it here. Is there anything special about your network setup? E.g. forced proxies or something of the kind? Greetings, Joachim -- Joachim nomeata Breitner Debian Developer nome...@debian.org | ICQ# 74513189 | GPG-Keyid: 4743206C JID: nome...@joachim-breitner.de | http://people.debian.org/~nomeata
