Dear R-devel On my Windows 2000 machine, I can build R-patched but not R-devel (since at least 1/9/2004). An error message pops up during the execution of some R code during "make", as below. In terms of "make" etc., I am totally clueless (not being a user of either Linux or C) and just mechanically follow the instructions in the R documentation; hence I've got no idea what to do when things snarf. More information below. Any advice appreciated-- I'm trying to test a patch for R2.0.0. Mark Background factoids: 1. The tools.zip file that I use is dated 9/5/2003. 2. I did expunge the R-devel directory before unzipping the .tar.gz and building. 3. I made my usual 2 changes to default MkRules (WINHELP=NO and TCL_HOME=C:/R/Tcl). 4. I don't have the Latex "bin" directory in the search path, contrary to src/gnuwin32/INSTALL-- but this hasn't been a problem in the past. With R-patched (and all the previous R versions I've built), I do get some errors/warnings about pdflatex not found but the build continues and the .exe files get created just fine. Here's what happens leading up to the error: C:\R\R-devel\src\gnuwin32> make << no worrying messages>> making bootstrap versions of packages ... -- base -- tools making DLL ... << no worrying messages >> ... DLL made installing DLL ... done cp ../../doc/html/R.css ../../library/R.css writing help indices for package: base tools utils grDevices graphics stats datasets methods grid sp lines stats4 tcltk ---------- Making package base ------------ adding build stamp to DESCRIPTION Error in get(x, envir, mode, inherits) : variable ".install_package_description" was not found Execution halted make[4]: *** [frontmatter] Error 1 make[3]: *** [all] Error 2 make[2]: *** [pkg-base] Error 2 make[1]: *** [rpackage] Error 2 make: *** [all] Error 2 ******************************* Mark Bravington CSIRO (CMIS) PO Box 1538 Castray Esplanade Hobart TAS 7001 phone (61) 3 6232 5118 fax (61) 3 6232 5012 Mark.Bravington@csiro.au
On Wed, 8 Sep 2004 11:01:25 +1000, <Mark.Bravington@csiro.au> wrote:>Dear R-devel > >On my Windows 2000 machine, I can build R-patched but not R-devel (since at least 1/9/2004). An error message pops up during the execution of some R code during "make", as below. In terms of "make" etc., I am totally clueless (not being a user of either Linux or C) and just mechanically follow the instructions in the R documentation; hence I've got no idea what to do when things snarf. > >More information below. Any advice appreciated-- I'm trying to test a patch for R2.0.0.I haven't seen the error you report, but the general strategy on seeing a make error is to "make distclean", and try again. If that doesn't work, it might be worth checking your Perl version: R 2.0.0 seems to need 5.8.x. Previously this was just recommended, and I ran on 5.6.x until very recently. I thought I had updated the Rtools page (www.murdoch-sutherland.com/Rtools) to say this, but looking just now shows the old page... Duncan Murdoch
Hi Duncan Thanks but... that doesn't seem to be it. I've upgraded Perl to 5.8 and the problem is still there. The error seems to be this: the "make" operation in R2.0.0 wants to use the *just-built* R.exe in the search path when it gets to the "Making package base" stage. But unless the user has modified their search path to include R-DEVEL\BIN before calling "make" (and made sure this comes before any other R BIN directories), the R.exe won't be found (on my Windows 2000 machine, anyway). I encountered the following: When I set PATH to start with C:\R\R-devel\bin, "make" works OK. When I leave PATH at its usual value that starts with C:\R\R-1.9.0\bin, "make" gives me this: ---------- Making package base ------------ adding build stamp to DESCRIPTION Error in get(x, envir, mode, inherits) : variable ".install_package_description" was not found Execution halted When I rename C:\R\R-1.9.0 so that there's no copy of R.exe in the PATH, "make" gave me this: ---------- Making package base ------------ adding build stamp to DESCRIPTION R: not found When I set PATH to start with a dot (looking at the INSTALL file section on "Check your path"), "make" fails again with the ".install_package_description" error. So it looks like either (i) the documentation should be changed to tell the user to reset the path manually (and NB the dot isn't useful for Windows 2000), or preferably (ii) "make" needs to temporarily change the search path, or to look more explicitly for the new R.exe file. Or I've gotten my wires crossed... Mark ******************************* Mark Bravington CSIRO (CMIS) PO Box 1538 Castray Esplanade Hobart TAS 7001 phone (61) 3 6232 5118 fax (61) 3 6232 5012 Mark.Bravington@csiro.au #-----Original Message----- #From: Duncan Murdoch [mailto:murdoch@stats.uwo.ca] #Sent: Wednesday, 8 September 2004 12:25 PM #To: Bravington, Mark (CMIS, Hobart) #Cc: r-devel@stat.math.ethz.ch #Subject: Re: [Rd] having trouble building r-devel # # #On Wed, 8 Sep 2004 11:01:25 +1000, <Mark.Bravington@csiro.au> wrote: # #>Dear R-devel #> #>On my Windows 2000 machine, I can build R-patched but not #R-devel (since at least 1/9/2004). An error message pops up #during the execution of some R code during "make", as below. #In terms of "make" etc., I am totally clueless (not being a #user of either Linux or C) and just mechanically follow the #instructions in the R documentation; hence I've got no idea #what to do when things snarf. #> #>More information below. Any advice appreciated-- I'm trying #to test a patch for R2.0.0. # #I haven't seen the error you report, but the general strategy on #seeing a make error is to "make distclean", and try again. # #If that doesn't work, it might be worth checking your Perl version: R #2.0.0 seems to need 5.8.x. Previously this was just recommended, and #I ran on 5.6.x until very recently. I thought I had updated the #Rtools page (www.murdoch-sutherland.com/Rtools) to say this, but #looking just now shows the old page... # #Duncan Murdoch #
Aha-- thanks!>> Or I've >> gotten my wires crossed... > >The latter. Try commenting out > ># path (possibly full path) to same version of R on the host system ># R_EXE=R >^^ > >in MkRules. It's uncommented there from experiments with >cross-compiling. >Works for me.Yes, it's just worked for me too. Is it worth mentioning this in the INSTALL file after the sentence that says "IMPORTANT: edit MkRules to set the appropriate paths..."? [It may be that all my builds hitherto have actually invoked an earlier version of R than the one being built, but that this never mattered before.]> >> So it looks like either (i) the documentation should be >changed to tell >> the user to reset the path manually (and NB the dot isn't useful for >> Windows 2000), > >It was when I ran Windows 2000. What's wrong for you?Just that, at the moment of running "make", the dot refers to src\gnuwin32, which doesn't pick up the R.exe file. With the commenting-out above, "make" now works fine for me now whether or not PATH starts with a dot. [I thought MS-DOS implicitly put a dot in, i.e. doesn't it always look for EXEs and BATs in the current directory first?] I think I need to start the path with <<current-R-version>>\bin rather than dot.>-- >Brian D. Ripley, ripley@stats.ox.ac.uk >Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ >University of Oxford, Tel: +44 1865 272861 (self) >1 South Parks Road, +44 1865 272866 (PA) >Oxford OX1 3TG, UK Fax: +44 1865 272595 >******************************* Mark Bravington CSIRO (CMIS) PO Box 1538 Castray Esplanade Hobart TAS 7001 phone (61) 3 6232 5118 fax (61) 3 6232 5012 Mark.Bravington@csir