Tcl Source Code

View Ticket
Login
Ticket UUID: 3018050
Title: fcopy-from-starkit hangs on Windows
Type: Bug Created on: 2010-06-18 14:09:06
Submitter: Assigned to: andreas_kupries
Subsystem: 25. Channel System Severity:
Priority: 8 Last modified: 2026-10-12 16:05:47
Status: Pending Closed by:
Resolution: Works For Me Closed on:
Version: obsolete: 8.5.8
Description:
Environment:
- Windows XP
- tclkit 8.5.8
- tcllhttpd 3.5.1 deployed as a starkit, with docRoot inside the starkit

When some files are requested, the response hangs, waiting for fcopy callback invocation. 

Steps to reproduce:
- launch the tclhttpd kit on Windows
- load the default page

You'll notice that /images/Space.gif request never completes

Other info
- It works fine on Mac OS X
- It works fine on Windows the doc root is outside the starkit
- http://groups.google.com/group/comp.lang.tcl/browse_thread/thread/261837ba204fb570#
User Comments:
ferrieux added on 2010-12-14 14:38:30:
That's fishy. Upping the prio.

cjmcdonald added on 2010-12-15 17:59:08:
This info may be relevant.

I encountered the same problem with another in-house virtual file
system implementation - not a starkit, but based on zip archives.  It
was serving out files using a httpd server, and this stopped
working.  I think it was when the 780533 fix was applied in 8.4.19.

On investigation I discovered that the binary mode fcopy now required
working fileevents.  This was not implemented in my zip file system,
which I took to be a deficiency there.  So I added a channel driver
Tcl_DriverWatchProc and associated mechanism for posting (if
requested) a file readable event after every read or seek - zip
archive files are always readable.  The fcopy then worked.

ferrieux added on 2010-12-15 18:41:49:
When you say that 'binary mode fcopy requires working fileevents',

- if you mean async fcopy (ie with -command), then yes, of course. In this mode fcopy is an alternating "pump" of readable/writable fileevents.

- if you mean sync fcopy (no -command), then no. In this mode fcopy does blocking IO, without resorting to the event loop.

Now for the starkit vfs, the async part works, on Linux. It's just a Windows issue apparently.

ferrieux added on 2011-08-19 03:35:33:
Is that still happening in 8.5.10 and 8.6b2 ?

serhiy.storchaka added on 2026-10-12 16:05:47:

I cannot reproduce it. Async fcopy from a reflected channel (as tclvfs uses) to a socket works in 8.5.8, 8.6.18 and 9.0, on Windows and Linux, if the channel posts readable events while it has data. If the driver does not post them, the copy stalls on all platforms, so this was most likely a problem in the VFS channel driver, as in the comment above.

Async fcopy from zipfs hangs for the same reason, see [0654cecae1].