Date: Wed, 18 Mar 2015 22:22:08 +0100 From: Ivan Radovanovic <radovanovic@gmail.com> To: freebsd-mono@freebsd.org Subject: Re: ASP.Net and file changes on FreeBSD Message-ID: <5509EC80.1060105@gmail.com> In-Reply-To: <5509E8B8.3090602@gmail.com> References: <54F96252.9040705@gmail.com> <55095322.2070604@gmail.com> <5509E8B8.3090602@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 03/18/15 22:06, Ivan Radovanovic napisa: > On 03/18/15 11:27, Ivan Radovanovic napisa: >> On 03/06/15 09:16, Ivan Radovanovic napisa: >>> Hi, >>> >>> I am running latest ASP.Net using latest mono from ports (3.10) on >>> FreeBSD 10.1 and it seems it is unable to detect changes on (some?) >>> files. What confuses me is that if I run xsp for testing project it is >>> unable to realize that project DLL is recompiled, or if I change >>> web.config it won't restart application manually, or if aspx file is >>> changed it won't recompile it, but if I add new aspx file then it will >>> try to compile that. It seems to me like it is able to detect file >>> addition but not change. >>> >>> Is there some workaround for this? >>> >>> Kind regards, >>> Ivan >>> >>> P.S. >>> Is there some dedicated website/forum for mono on FreeBSD - I saw some >>> page on mono project site but that is completely outdated and I also saw >>> BSD# page on google, but that also seems little bit old? I did little >>> bit of work on having up-to-date xsp/fast-cgi and monodevelop to >>> compile/work on freebsd so I would like to pass that to somebody from >>> freebsd-mono group so it could eventually reach ports :-) >> >> This is actually more interesting problem than what I expected, here are >> conclusions up to this point: >> >> * 3.10.0 has completely broken kevent implentation of FileSystemWatcher >> - so it couldn't restart application because it is not able to detect >> any file changes (I gave up on patching this implementation with latest >> code when I saw that 3.12.1 is ported to FreeBSD) >> >> * 3.12.1 has proper support for FileSystemWatcher, but I suspect >> something in System.Web.dll is broken (for example it properly detects >> dll change (writes message about that), but doesn't restart >> application), I am still unable to pinpoint exact problem there >> >> * 3.2.3 has partially working FileSystemWatcher (using kevent) - it is >> not able to catch file changes but only file creations/deletions, for >> that reason changes using editors like ee which really change file >> contents (not changing inode number) go undetected, but changes with >> editors which recreate file (like monodevelop or gedit for example) are >> properly captured and handled. This version also properly restarts >> application when dll is changed. >> >> I was using identical xsp for all mono versions to rule out differences >> there (I didn't do any tests using fastcgi server or mod_mono - will >> probably do that once I have xsp working as expected). > > There were 2 problems here (applies to 3.12.1): > > * FileSystemWatcher implementation for kevent was relying on Darwin's > specific behavior (when kevent is executed with NULL timeout, if kqueue > FD is closed kevent call on Darwin returns with error, however FreeBSD > doesn't behave that way), so instead of using (almost) blocking version > of kevent I put reasonable timeout there (instead of 500 seconds they > had I put 0.5 seconds) > > * HttpApplicationFactory wasn't able to restart application domain > because of deadlock within FileSystemWatcher (basically file changed > notification is executed from watcher's thread and then there is attempt > to disable watcher from that same thread). I resolved this by creating > separate thread to post events. > > As far as I can tell at least xsp now works as expected (restaring > application on DLL changes, recompiling changed files if .aspx file is > changed), I will probably test at least fastcgi server soon as well, and > I will try to find solution which would work with blocking kevent. > > Attached is patch for latest mono version in ports (3.12.1) - patch file > should be saved to files subdirectory of port and port should be rebuilt > after that. Previous patch had one mistake (timespec struct in fact uses nanoseconds, not microseconds (mono guys named member tv_usec, so I naturally assumed microseconds were used)) - I realized that when I checked CPU usage. This is final patch attached.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5509EC80.1060105>