diff options
-rw-r--r-- | NEWS | 4 | ||||
-rw-r--r-- | www/upstream-bugs.html | 14 |
2 files changed, 17 insertions, 1 deletions
@@ -1,4 +1,5 @@ * Sat May 8 2010 Eric S. Raymond <esr@snark.thyrsus.com> - 2.95~dev + The autonomous robot submarine total world domination release! Rationalize clearing and generation of DOPs, this makes epx/epy much more generally available. Fixed the test productions for the udev magic and added a troubleshooting note in INSTALL. cgps now displays @@ -12,7 +13,8 @@ Designated Area Code and Functional ID pair. gpspipe has a new -T option for setting the timestamp format. xgpsspeed is completely rewritten in Python, eliminating some dependencies on ancient X - libraries. We now ship a Qt binding for the client library. + libraries. We now ship a Qt binding for the client library. Note + a GCC 4.2.1 optimizer bug. * Tue Apr 20 2010 Eric S. Raymond <esr@snark.thyrsus.com> - 2.94 Error-checking in the 50bps subframe code has been greatly improved. diff --git a/www/upstream-bugs.html b/www/upstream-bugs.html index 2679f47a..8b16c2b0 100644 --- a/www/upstream-bugs.html +++ b/www/upstream-bugs.html @@ -74,6 +74,7 @@ maintainer, we've tried to kick these upstream.</p> <ul> <li><a href="#libtool">Multiple definitions of symbol _gpsd_report due to libtool bug</a></li> <li><a href="#pyinstall">Defective Python installation in SuSE 10.3rc</a></li> +<li><a href="#gcc421">Compiler optimizer bugs can produce error-modeling errors<a></li> </ul> <h1>Links to Fixed Bugs</h1> @@ -202,6 +203,19 @@ in.</p> Makefile. This prevents the <code>gpsd</code> Python components from building correctly.</p> +<h2 id="gcc421">Compiler optimizer bugs can produce error-modeling errors<</h2> + +<p>The floating-point code in the daemon's error modeler (the logic +that produces uncertainty estimates in the sensor doesn't supply them) +is sensitive to optimizer bugs. If the optimizer is flaky, different +levels of operation can cause the speed calculation to either converge +or blow up in FP operations; the effect is that when it is actually +reported seems to vary randomly with level of optimization.</p> + +<p>Under GCC 4.2.1, -O2 diverges from what GCC 4.4.1 reports, causing +regression-test failures. Dropping back to -O1 or going up to -O3 +restores behavior like GCC 4.4.1's.</p> + <h1>Fixed Problems</h1> <p>We document these here in case you're running on an older system. |