Up to now static drivers were created via co65 from dynamic drivers. However there was an issue with that approach:
The dynamic drivers are "o65 simple files" which obligates that they start with the 'code' segment. However dynamic drivers need to start with the module header - which is written to. For dynamic drivers this isn't more than a conceptual issue because they are always contain a 'data' segment and may therefore only be loaded into writable memory.
However when dynamic drivers are converted to static drivers using co65 then that issue becomes a real problem as then the 'code' segment may end up in non-writable memory - and thus writing to the module header fails.
Instead of changing the way dynamic drivers work I opted to rather make static driver creation totally independent from dynamic drivers. This allows to place the module header in the 'data' segment (see 'module.mac').
After all it seems much cleaner to explicitly control the behaviour from the .travis.yml file than to behave differently "under the hood" when detecting a Travis CI build.
I must admit that I don't understand why but obviously it is necessary to place .PHONY below the definition of variables it refers to - although those variables are recursively expanded ones! Not doing so made libsrc/Makefile build only three target libraries.
Building the clean target in src or libsrc should only delete stuff created by the make in those directories. Having both separated allows the Travis CI build to replace the native binaries with cross built binaries while keeping everything else.
It seems more appropriate to trigger the inclusion of Makefile.inc via $TARGET (instead of $SRCDIR). This is btw. more consistent with extzp.s inclusion which is triggered via $TARGET too.
The 'install' target primarily aims to support pacaking tools. Therefore...
- It just presumes a "capable" install program to be present.
- There's intentionally no 'uninstall' target.
Originally I used the usual variables (like $(CC) and $(CFLAGS) ) but after
all this doesn't make sense as any predefined values and/or user defined
settings can only be wrong.
The libraries build is prepared to work with binaries not part of the current
working tree. But in this case the default search path surely points to some
other working tree. Even the binaries in this working tree might have been
compiled with non-standard builtin search paths.
Anyway when building the libraries we want always to use the headers from
the current working tree. Therefore we want to set them explicitly. Instead
of setting all in all five paths on the command lines of ca65, cc65 and ld65 I
opted to just set the single environment variable.
The target 'atarixl' is to be used for Atari XL (and better) machines.
It will disable the OS ROM and enable the Shadow RAM available on
those machine.
Note: This commit is only the inital step towards for this goal that just
replicates the target 'atari' as a starting point!
The targets allow to run cc65 programs in the sim65 exection
einvironment. As there are no "real" i/o facilities there's no
need for header files. Paravirtualized entry points are mapped
to $FFF0 ff. There's a large cc65 progam area from $0200-$FFEF.
The binary format includes a one-byte header indicating the required
execution environment: The value 0 means 6502 and the value 1
means 65C02. The load adress for the binary is fixed to $0200.
Note: Running sim65C02 programs currently doesn't work bcause
sim65 doesn't actually implement 65C02 opcodes.
There's a quite special case of a targetutil for the TARGET geos-apple having to be
built as TARGET apple2enh binary. We can't just add a dependency to ../lib/apple2enhlib
because the TARGETs are built in individual make instances (because of vpath).
Doing that the "right way" would be _very_ complicated and would turn the Makefiles way
less readable. Therefore I decided to solve that by just reordering the TARGETs. Surely
this won't work if i.e. 'make geos-apple' is entered on a clean system. But still I consider
it appropriate.
I ordered the TARGETs btw so that they are more in an alphabetical order...
Now that we're pretty much independent from CC65_HOME
it doesn't make sense anymore to look for it in the Makefiles.
So rather dynamically check if there are "local" binaries in the
../bin directory. If there are "local" binaries use them - and
otherwise rely on "installed" binaries found in the path.
- No complex shell logic.
- "Source file shadowing" for all targets via vpath.
- Dependency handling.
- True incremental build.
- Don't write into source directories.
- Easy cleanup by just removing 'wrk'.
Now the whole GEOSLib is built in a single make instance running in 'libsrc/geos-cbm' - which is just what the cc65 library build system can handle. The 'libsrc/geos-cbm' sub-directories still contain Makefiles. However those files only define the set of object files to create from their sub-directory and get included into the 'libsrc/geos-cbm' Makefile.
git-svn-id: svn://svn.cc65.org/cc65/trunk@5429 b7a2c559-68d2-44c3-8de9-860c34a00d81
These two patches speed up library building:
Do the "runtime" directory before doing the platform directory, because
platform drivers need a file from the runtime directory (the old code
ran "runtime/Makefile" three times for each target!!!).
Clean only the shared directories -- the target directories don't need
cleaning.
Changes by Greg King.
git-svn-id: svn://svn.cc65.org/cc65/trunk@5019 b7a2c559-68d2-44c3-8de9-860c34a00d81
