home *** CD-ROM | disk | FTP | other *** search
- Path: ix.netcom.com!news
- From: giuliano@ix.netcom.com(Giuliano Carlini)
- Newsgroups: comp.lang.c++
- Subject: Re: Slow longjmp in Watcom 10.0
- Date: 18 Apr 1996 20:31:29 GMT
- Organization: Netcom
- Message-ID: <4l68r1$7jc@dfw-ixnews6.ix.netcom.com>
- References: <4kvejg$ne9@goanna.cs.rmit.edu.au>
- NNTP-Posting-Host: lbx-ca5-09.ix.netcom.com
- X-NETCOM-Date: Thu Apr 18 3:31:29 PM CDT 1996
-
- In <4kvejg$ne9@goanna.cs.rmit.edu.au> jtai@numbat.cs.rmit.EDU.AU (Jason
- Chen Hiung Tai) writes:
- >
- >I did a test of longjmp under OS/2 using Watcom C/C++ 10.0 on
- >a P90. What shock me was that it took almost 2400 cycles or ...
-
- It's been a while since I messed with longjmp with Watcom, so take this
- with a boulder of salt...
-
- If I remember correctly, Watcom's longjmp is unwinding some runtime
- data structures. These data structures are normally used to call the
- destructors for C++ objects on the stack. longjmp has to unwind these,
- or the runtime gets really confused, and has a stroke. Borland has a
- flag to disable exception handling. Does Watcom have a similar flag. If
- so, it may make longjmp more efficient. The only way to tell for sure
- is to single step the longjmp code in the debugger. Painful, but
- necessary.
-
- g
-