Tcl Source Code

 The demo script reads part of a file, then changes
the buffer size, then tries to read the rest of the
file.  Then, either a segfault or a POSIX error.

To diagnose the memory problem, evaluate the demo
script in a tclsh built with TCL_MEM_DEBUG:

$ ./tclsh buffer.tcl 
hi guard byte 0 is 0x65         e
hi guard byte 2 is 0x64         d
hi guard byte 3 is 0x20          
hi guard byte 4 is 0x24         $
hi guard byte 5 is 0x63         c
total mallocs                   1402
total frees                      791
current packets allocated        611
current bytes allocated        49654
maximum packets allocated        759
maximum bytes allocated        87378
high guard failed at 80d9008, ./../generic/tclIO.c 4913
58 bytes allocated at (./../generic/tclIO.c 1659)
Memory validation failure
Aborted (core dumped)

Apparently this problem first appeared in Tcl 8.1.

Logged In: YES 
user_id=75003

Performance patch committed to main and core-8-3-1-branch.
This item is now completely closed.

File Added - 8533: tcl.427196.2.diff.txt

Logged In: YES 
user_id=75003

Upload of patch which squashes the performance problem
for dynamically changing buffersizes, by squashing buffers
which are to small.

File Added - 8526: tcl.427196.1.diff.txt

File Added - 8523: tcl.427196.0.diff.txt

Logged In: YES 
user_id=75003

The reason for the crash is a combination of two things: 
First, the buffer allocated when -buffersize 10 was set is 
reused in the second read. And second, the IO system 
blithely uses the set buffersize as the number of bytes to 
read and not the actual size of the buffer in use. This 
causes the overwrite beyond the end of the buffer and the 
subsequent crash when closing the channel.

The relevant location is tclIO.c, line 5004 in "GetInput".

A patch will follow shortly.

File Added - 6648: buffer.tcl