Jack, On Sonntag, 17. Juni 2018 16:55:06 CEST Jack wrote:
> On 2018.06.16 02:13, Thomas Baumgart wrote: > > On Samstag, 16. Juni 2018 00:45:38 CEST Jack Ostroff wrote: > > > >> Recently, I've been playing with some informal comparisons between > >> make and ninja as build system, and between gcc and clang as > >> compiler. I have a sense that ninja is a bit faster, although I've > >> also been trying cccache, which make a great difference if doing > >> clean compiles but with only minor changes in configure parameters. > > > > >> However, starting about two days ago, several builds ended up not > >> having libofx clientuid capability, which is absolutely essential > >> for me, and which has worked fine for quite a long time. I started > >> digging into the code, and I did notice two recent commits in CMake > >> files related to libofx. However, nothing made any sense, until I > >> realized that what happened was that all my compiles using "make" > >> failed to find clientuid support, but my compiles with "ninja" did > >> find it. Note that terminology is not quite accurate, since the > >> success or failure was all within the cmake/ccmake run, before > >> either make or ninja was involved. I'm guessing this is a bug within > >> cmake itself, but I'm stumped as to how to create a small test case. > > > > >> Question: is there anyone compiling from source who has found the > >> problem with cmake not finding clientuid support, where it was > >> previously found? I'm really not sure if one of the cmake related > >> commits tripped over some sort of edge case, > > > > >> I did reinstall cmake on 23 May, but it was just a reinstall of the > >> same version 3.9.6 I originally installed on 27 Dec. I'm going to > >> try a newer version of cmake, but would love to find I'm not the > >> only one who has gotten caught by this. > > > > > > Thanks for any ideas. > > > > > > Jack > > > > Can you check if the attached patch makes a difference, please? > > Yes, it works. Thanks. I added it to master. > Now I'm curious why it worked when preparing for Ninja but not for > make. I also don't seem to follow the logic of why it didn't find the > clientuid support, but DID find libofx. Because the variable LIBOFX_INCLUDE_DIR was not set (only LIBOFX_INCLUDE_DIRS was by PKG_CHECK_MODULES). So it checked for the clientuid structure member in a file /libofx/libofx.h which was not found. Why it worked for the ninja generator? I don't know. This is cmake magic. I also don't know, why for FreeBSD you need to have PC_OFX instead of LIBOFX. This is another mistery to me. Thomas -- Regards Thomas Baumgart https://www.telegram.org/ Telegram, the better WhatsApp ------------------------------------------------------------- Memory's the second thing to go ... Can't remember the first. -------------------------------------------------------------
signature.asc
Description: This is a digitally signed message part.
