On Mon, Jun 11, 2012 at 02:29:59AM -0700, Thierry Reding wrote:
PGP Signed by an unknown key
Stephen Warren wrote:
Allen Martin amartin@nvidia.com wrote:
On Fri, Jun 08, 2012 at 10:28:57PM -0700, Stephen Warren wrote:
On 06/08/2012 03:27 PM, Allen Martin wrote:
On Fri, Jun 08, 2012 at 02:07:25PM -0700, Stephen Warren wrote:
On 06/08/2012 02:01 PM, Allen Martin wrote: > On Fri, May 25, 2012 at 06:46:22AM -0700, Thierry Reding wrote: >> This commit uses the common Tegra board implementation instead
of
>> duplicating a lot of the code. In addition, the Plutux and
Medcom
>> specific board files can be removed as the MMC/SD setup is
common
>> among all Tamonten-based boards. >> >> ... >> diff --git a/board/avionic-design/medcom/Makefile
b/board/avionic-design/medcom/Makefile
>> index b0c318c..d96d043 100644 >> --- a/board/avionic-design/medcom/Makefile >> +++ b/board/avionic-design/medcom/Makefile >> @@ -26,12 +26,12 @@ >> include $(TOPDIR)/config.mk >> >> ifneq ($(OBJTREE),$(SRCTREE)) >> -$(shell mkdir -p $(obj)../common) >> +$(shell mkdir -p $(obj)../common $(obj)../../nvidia/common) >> endif >> >> LIB = $(obj)lib$(BOARD).o >> > > This breaks with my patchset to split out the arm7 code to an SPL > because even though SRCDIR and OBJDIR are the same, in the SPL
build
> $(obj) is a subdir of $(SPLTREE) not $(OBJTREE) (not my doing). > > How about the following instead which seems more to the point: > > ifeq ($(wildcard $(obj)../common),) > $(shell mkdir -p $(obj)../common) > endif > ifeq ($(wildcard $(obj)../../nvidia/common),) > $(shell mkdir -p $(obj)../../nvidia/common) > endif
Maybe I'm just not reading it right, but isn't that just running
the
exact same mkdir commands, just splitting it into two commands,
and
making them optional based on the $(wildcard)? I'm still not clear
what
the problem is.
Anyway, if this is an issue, then compal/paz00 and
compulabl/trimslice
will need to be fixed for the SPL changes in the same way.
The problem is this:
>> ifneq ($(OBJTREE),$(SRCTREE))
is true, but these directories:
>> +$(shell mkdir -p $(obj)../common $(obj)../../nvidia/common)
don't exist because $(obj) is not under $(OBJTREE) in the SPL
build,
it's under $(SPLTREE)
OK, so I think the issue is just that $(obj) isn't the correct place
to
mkdir these, not that we shouldn't attempt to mkdir any directory. In other words, don't you need to switch between mkdir in the object
tree
or the SPL tree rather than switching between doing a mkdir and not? It's quite possible this only affects builds where OBJDIR!=SRCDIR, so
a
regular build might not show up the problem I /think/ exists in the
code
quoted a couple of emails above.
$(obj) is always the place where the object files are going to be placed, that may be in $(OBJTREE) or in $(SRCTREE) or in $(SPLTREE) depending on what you're building and what you have set.
I think the $(wildcard) suggestion I made is the most straightforward, it basically says "we're about to spit out an object file to here, so make sure the directory exists first". In the "normal" case where your object files are in your source tree the directory will already exist so it won't do anything.
Ah right, I understand now.
But why not make it super-simple and just run the mkdir unconditionally; it won't cause any harm for an in-tree object location, but removes the need for anyone to think about the conditional logic.
Shouldn't this really be handled by the commands that create files in those directories? They could check that the directories exist, and create them if necessary, before executing their actual commands.
The normal rules make sure that $(obj) exists, it's only because these boards generate object files in $(obj)../common and $(obj)../../nvidia/common that there's a problem and we need to do something special.
Stephen's idea of just the unconditional mkdir makes the most sense, we don't really save anything by skipping it if the directory already exists.
-Allen