This release may cause trouble. It has been found that an application (QGIS) plugin (SDA4P) loading R and rgdal through rpy2, and with QGIS linking to GDAL itself directly, has been occasionally seeing muddles with regard to error handling on Linux. The CRAN Windows and CRAN extras OSX binary packages are built static, so don't see the problem (they have private copies of GDAL), but in order to try to resolve this, and following exchanges with Even Rouault, a GDAL developer, major changes in error handling in rgdal have been made. Before, the R error handler was set to catch GDAL errors on the R thread (GDAL's error handler is said to be threaded, so rgdal pushed the R handler on load and popped it on unload). If the plugin loaded R on the same thread as QGIS, the R error handler might be seen by QGIS, instead of the QGIS error handler. Now, the error handler is pushed/popped around every call to a GDAL function, on occasion pushed before a loop and popped after the loop. I have contacted all maintainers of packages on CRAN depending on or suggesting rgdal, so that they could check whether their packages are affected. William Kyngesburye has been kind enough to clear some other issues for his R 2.13 Framework build, but reports no problems so far. If you see issues, in particular R error exiting because the wrong error handler is being applied (only Linux, OSX with rgdal intalled from source or Frameworks, or Windows with rgdal installed from source), please document fully, preferably reproducibly, and report, and, until fixed, revert to the 0.6-* series. Windows and OSX binaries will not be built for GDAL 1.8.0, but we'll aim to build with GDAL 1.8.1 sometime after its (impending) release. All feedback welcome, including scrutiny of the changes made in the C/C++ source code on R-Forge. Roger
Roger Bivand Department of Economics, NHH Norwegian School of Economics, Helleveien 30, N-5045 Bergen, Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43 e-mail: Roger.Bivand at nhh.no