My makefile failed. My idea was to extract files from a certain file and put them in a temporary directory. This is then rsync'ed to a source directory. Since rsync only updates when a file has changed, makefiles in the source directory do not remake unless necessary. I have isolated the problem below:
.PHONY: extract clean
FILES = filea1.cc filea2.cc filea3.cc \
filea1.hh filea2.hh filea3.hh \
filea1.py filea2.py filea3.py \
dir-cc/filea1.cc dir-cc/filea2.cc dir-cc/filea3.cc \
dir-cc/filea1.hh dir-cc/filea2.hh dir-cc/filea3.hh \
dir-py/filea1.py dir-py/filea2.py dir-py/filea3.py
PATTERN := $(sort $(addprefix build/%., $(patsubst .%,%,$(suffix $(FILES)))))
extract: $(addprefix build/, $(FILES))
$(PATTERN):
mkdir -p $(dir $#); echo "hello!" > $#
#build/%.cc:;mkdir -p $(dir $#); echo "hello!" > $#
#build/%.hh:;mkdir -p $(dir $#); echo "hello!" > $#
#build/%.py:;mkdir -p $(dir $#); echo "hello!" > $#
temp:
echo $(PATTERN)
clean:
rm -rf build
Doing make temp gets 'build/%.cc build/%.hh build/%.py' but 'make extract' fails to make all the files:
build:
dir-cc dir-py filea1.cc filea2.cc filea3.cc
build/dir-cc:
filea1.cc filea2.cc filea3.cc
build/dir-py:
filea1.py filea2.py filea3.py
The hh files are missing. Unbelievably, in the actual work, the cc files are missing. Anyway, commenting out $(PATTERN), and removing the comments so that the targets in the code above are separate, does get all the files:
build:
dir-cc filea1.cc filea1.py filea2.hh filea3.cc filea3.py
dir-py filea1.hh filea2.cc filea2.py filea3.hh
build/dir-cc:
filea1.cc filea1.hh filea2.cc filea2.hh filea3.cc filea3.hh
build/dir-py:
filea1.py filea2.py filea3.py
Since everything has to be automatic in the actual work, and the rule is the same for all the targets (so I should not have to retype it), I vastly prefer the first version. I read the manual but I am perplexed as to why make does this. I am using GNU Make 4.1.
It's explained in the last paragraph of the Pattern Intro section:
Pattern rules may have more than one target. Unlike normal rules, this does not act as many different rules with the same prerequisites and recipe. If a pattern rule has multiple targets, make knows that the rule’s recipe is responsible for making all of the targets. The recipe is executed only once to make all the targets.
You can't write a pattern rule with many different pattern targets, such that make will run the recipe once per individual target. That's not how it works.
You could do this, for your very simple case:
FILETYPES := $(sort $(suffix $(FILES)))
$(foreach T,$(FILETYPES),$(eval build/%$T: ; mkdir -p $$(#D); echo "hello!" > $$#))
There was a good resolution: use the ordinary (non pattern) multiple target
rule $(FILES):;rule and define pattern specific variable values to adjust
the rule to the file extension.
Related
I am trying to statically build HTML files that requires a markdown file and a meta file called "whatlinkshere" for the HTML file to demonstrate its back links.
I believe it can be effeciently done by a Makefile, by first generating all the "whatlinkshere" files. I don't think this can be done in parallel, because the program that generates these files needs to append to the whatlinkshere files, and there could be race conditions that I am not quite sure how to solve.
Once the "whatlinkshere" files are generated then if a markdown file is edited, say foo.mdwn to point to bar.mdwn, only foo.mdwn needs to be analysed again for "whatlinkshere" changes. And finally only foo.html and bar.html need to be rebuilt.
I am struggling to accomplish this in my backlinks project.
INFILES = $(shell find . -name "*.mdwn")
OUTFILES = $(INFILES:.mdwn=.html)
LINKFILES = $(INFILES:.mdwn=.whatlinkshere)
all: $(OUTFILES)
# These need to be all made before the HTML is processed
$(LINKFILES): $(INFILES)
#echo Creating backlinks $#
#touch $#
#go run backlinks.go $<
%.html: %.mdwn %.whatlinkshere
#echo Deps $^
#cmark $^ > $#
Current problems here is that *.whatlinkshere** aren't being generated on first run. My workaround is for i in *.mdwn; do go run backlinks.go $i; done. Furthermore there are not rebuilding as I want after editing a file as described earlier. Something is horribly wrong. What am I missing?
I think I finally understood your problem. If I understood well:
You have a bunch of *.mdwn source files.
You generate *.whatlinkshere files from your *.mdwn source files using the backlinks.go utility. But this utility does not produce foo.whatlinkshere from foo.mdwn. It analyzes foo.mdwn, searches for links to other pages in it and, for each link to bar it finds, it appends a [foo](foo.html) reference to bar.whatlinkshere.
From each foo.mdwn source file you want to build a corresponding foo.html file with:
$ cmark foo.mdwn foo.whatlinkshere
Your rule:
$(LINKFILES): $(INFILES)
#echo Creating backlinks $#
#touch $#
#go run backlinks.go $<
contains one error and has several drawbacks. The error is the use of the $< automatic variable in the recipe. It expands as the first prerequisite, that is probably always pageA.mdwn in your case. Not what you want. $^ expands as all prerequisites but it is not the correct solution because:
your go utility takes only one source file name, but even if it was accepting several...
...make will run the recipe several times, one per link file, which is a waste, and...
...as your go utility appends to the link files it will even be worse than a waste: back links will be counted several times each, and...
...if make runs in parallel mode (note that you can prevent this with make -j1 or by adding the .NOTPARALLEL: special rule to your Makefile, but it is a pity) there is a risk of race conditions.
Important: the following works only with a flat organization where all source files and HTML files are in the same directory as the Makefile. Other organizations are possible, of course, but they would require some modifications.
First option using multi-targets pattern rules
One possibility is to use a special property of make pattern rules: when they have several targets make considers that one single execution of the recipe produces all targets. For instance:
pageA.w%e pageB.w%e pageC.w%e: pageA.mdwn pageB.mdwn pageC.mdwn
for m in $^; do go run backlinks.go $$m; done
tells make that pageA.whatlinkshere, pageB.whatlinkshere and pageC.whatlinkshere are all generated by one execution of:
for m in pageA.mdwn pageB.mdwn pageC.mdwn; do go run backlinks.go $m; done
(make expands $^ as all prerequisites and $$m as $m). Of course, we want to automate the computation of the pageA.w%e pageB.w%e pageC.w%e pattern targets list. This should make it:
INFILES := $(shell find . -name "*.mdwn")
OUTFILES := $(INFILES:.mdwn=.html)
LINKFILES := $(INFILES:.mdwn=.whatlinkshere)
LINKPATTERN := $(INFILES:.mdwn=.w%e)
.PHONY: all clean
.PRECIOUS: $(LINKFILES)
all: $(OUTFILES)
# These need to be all made before the HTML is processed
$(LINKPATTERN): $(INFILES)
#echo Creating backlinks
#rm -f $(LINKFILES)
#touch $(LINKFILES)
#for m in $^; do go run backlinks.go $$m; done
%.html: %.mdwn %.whatlinkshere
#echo Deps $^
#cmark $^ > $#
clean:
rm -f $(LINKFILES) $(OUTFILES)
Notes:
I declared all and clean as phony because... it is what they are.
I declared the whatlinkshere files as precious because (some of them) are considered by make as intermediates and without this declaration make would delete them after building the HTML files.
In the recipe for the whatlinkshere files I added rm -f $(LINKFILES) such that, if the recipe is executed, we restart from a clean state instead of concatenating new stuff to old (possibly outdated) references.
The pattern stem in the $(LINKPATTERN) can be anything but must match at least one character. I used w%e but whatlin%shere would work too. Use whatever is specific enough in your case. If you have a pageB.where file prefer whatlin%shere or what%here.
There is a drawback with this solution but it is due to your particular set-up: each time one single mdwn file changes it must be re-analyzed (which is normal) but any whatlinkshere file can be impacted. This is not predictable, it depends on the links that have been modified in this source file. But more problematic is the fact that the result of this analysis is appended to the impacted whatlinkshere files. They are not "edited" with the old content relative to this source file replaced by the new one. So, if you change just a comment in a source file, all its links will be appended again to the respective whatlinkshere files (while they are already there). This is probably not what you want.
This is why the solution above deletes all whatlinkshere files and re-analyzes all source files each time one single source file changes. And another negative consequence is that all HTML files must also be re-generated because all whatlinkshere files changed (even if their content did not really change, but make does not know this). If the analysis is super fast and you have a small number of mdwn files, it should be OK. Else it is sub-optimal but not easy to solve because of your particular set-up.
Second option using recursive make, separated back link files and marker files
There is a possibility, however, which consists in:
separating all back links references with one whatlinkshere file per from/to pair: foo.backlinks/bar.whatlinkshere contains all references to bar found in foo.mdwn,
using recursive make with one first invocation (when the STEP make variable is unset) to update all whatlinkshere files that need to be and a second invocation (STEP set to 2) to generate the HTML files that need to be,
using empty dummy files to mark that a foo.mdwn file has been analyzed: foo.backlinks/.done,
using the secondary expansion to be able to refer to the stem of a pattern rule in its list of prerequisites (and using $$ to escape the fist expansion).
But it is probably a bit more difficult to understand (and maintain).
INFILES := $(shell find . -name "*.mdwn")
OUTFILES := $(INFILES:.mdwn=.html)
DONEFILES := $(patsubst %.mdwn,%.backlinks/.done,$(INFILES))
.PHONY: all clean
ifeq ($(STEP),)
all $(OUTFILES): $(DONEFILES)
$(MAKE) STEP=2 $#
%.backlinks/.done: %.mdwn
rm -rf $(dir $#)
mkdir -p $(dir $#)
cp $< $(dir $#)
cd $(dir $#); go run ../backlinks.go $<; rm $<
touch $#
else
all: $(OUTFILES)
.SECONDEXPANSION:
%.html: %.mdwn $$(wildcard *.backlinks/$$*.whatlinkshere)
#echo Deps $^
#cmark $^ > $#
endif
clean:
rm -rf *.backlinks $(OUTFILES)
Even if it looks more complicated there are a few advantages with this version:
only outdated targets are rebuilt and only once each,
all whatlinkshere files are updated (if needed) before any HTML file is updated (if needed),
the whatlinkshere files can be built in parallel,
the HTML files can be built in parallel.
Third option using only recursive make and marker files
If you do not care about inaccurate results where back links persist in the results after they disappeared from the source files or where back links are uselessly replicated, we can reuse ideas from the previous solution but drop the separation in individual from/to whatlinkshere files.
INFILES := $(wildcard *.mdwn)
OUTFILES := $(patsubst %.mdwn,%.html,$(INFILES))
LINKFILES := $(patsubst %.mdwn,%.whatlinkshere,$(INFILES))
DONEFILES := $(patsubst %.mdwn,.%.done,$(INFILES))
.PHONY: all clean
.PRECIOUS: $(LINKFILES)
ifeq ($(STEP),)
.NOTPARALLEL:
all $(OUTFILES): $(DONEFILES)
$(MAKE) STEP=2 $#
.%.done: %.mdwn
go run backlinks.go $<
touch $#
else
all: $(OUTFILES)
%.html: %.mdwn %.whatlinkshere
#echo Deps $^
#cmark $^ > $#
%.whatlinkshere:
touch $#
endif
clean:
rm -f $(OUTFILES) $(LINKFILES) $(DONEFILES)
Notes:
As this works only for a flat organization I replaced the $(shell find...) by the make built-in $(wildcard ...).
I used patsubst instead of the old syntax but it's just a matter of taste.
The %.whatlinkshere: rule is a default rule to create the missing empty whatlinkshere files.
The NOTPARALLEL: special target prevents parallel execution when building the whatlinkshere files.
I have a command in Makefile:
# just a list of all package names separated by newline:
# gomq/consumers/aaa
# gomq/rock
PKGS=$(shell go list ./...)
# Coverage output dirs: coverage/$PKG/coverage.out
COVPKGS=$(addsuffix /coverage.out,$(addprefix coverage/,$(PKGS)))
final-command: coverage/all.out
some-commands
coverage/all.out: $(COVPKGS)
echo "mode: set" >$#
grep -hv "mode: set" $(wildcard $^) >>$#
coverage/gomq/consumers/%:
# mkdir -p $(dir $#)
# go test -coverprofile $# $(patsubst coverage/%/coverage.out,%,$#) -redisAddr=$(REDIS_SERVER)
coverage/gomq/%:
# mkdir -p $(dir $#)
# go test -coverprofile $# $(patsubst coverage/%/coverage.out,%,$#)
So I got a list of directories in $(COVPKGS), and I want to pattern match the list against coverage/gomq/consumers/% and coverage/gomq/% to fire different commands for different directories.
I think Make figures out those dependencies and tries to match them to a target so coverage/gomq/consumers/% will match consumer targets and coverage/gomq/% will match the rest.
But when I run it, it can only match coverage/gomq/whatever, cannot match any further down the directory like coverage/gomq/whatever/whatever or coverage/gomq/consumers/whatever. So some packages deep down in the directory cannot be tested. Any way to correct it? Thanks.
Actually, if the directories already exists with files, and I run make again, the target will not be matched.
first time here. I am relatively new to makefiles. Here is my current makefile:
# Closure compiler php script path
closure = ../../cli/scripts/Compilers/closure.php
# Destination directory
dest = ../../static/js/
# Build directory
build = build/
# Tell "make" to search build and destination dirs
vpath %.o $(build)
vpath %.js $(dest)
all: main.js
#echo "Done.";
main.js: \
date.o \
jquery.autocomplete.o \
jquery.bullseye.o \
jquery.clickopen.o \
jquery.fbmodal.o \
jquery.helpers.o \
jquery.pulljson.o \
jquery.thumbrotate.o \
jquery.timefmt.o \
jquery.tools.o \
layout.main.o
cat $^ > $(dest)$#
%.o: %.js
php $(closure) $*.js $(build)$#
clean:
rm -rf $(build)*.o
rm -rf $(dest)*.js
The problem is with the following line:
cat $^ > $(dest)$#.
It is supposed to cat all the prerequisite objects (minified javascript) into one final js library. According to makefile docs, $^ is an automatic variable which contains a list of prerequisites with directories they are in. From my experience, it behaves differently depending on wether prerequisite needs to be compiled or not.
If prerequisite is up-to-date, this code works perfectly and $^ contains a list like:
build/date.o build/jquery.autocomplete.o build/jquery.bullseye.o....
However if prerequisite needs a fresh compile, then $^ gets directory part stripped and looks like:
date.o jquery.autocomplete.o jquery.bullseye.o
Only the file which needs a fresh compile gets directory part stripped.
I have managed to work around this issue by replacing
cat $^ > $(dest)$#
with
cat $(addprefix $(build), $(^F) ) > $(dest)$#.
I don't like it because:
It's a hack
$(^F) is semi-deprecated
I want to understand why make behaves like this.
thanks
Look here:
# Tell "make" to search build and destination dirs
vpath %.o $(build)
If Make is looking for foo.o, it will look in the local directory first. If it finds no foo.o there, it will look in $(build) (i.e. build/, and you might reconsider your variable names).
And how would Make build foo.o, if it couldn't find it anywhere? With this rule:
%.o: %.js
php $(closure) $*.js $(build)$#
This rule violates an important guideline of makefiles, in that the target (foo.o) is not the name of the thing actually built (build/foo.o).
Now consider what happens when Make tries to execute this rule:
main.js: date.o ...
cat $^ > $(dest)$#
So if date.o is up to date, it's in build/. Make finds it there, and the automatic variable $^ expands to build/date.o ...
But if date.o must be rebuilt, then Make looks to the %.o rule, which promises to build date.o (not build/date.o), so Make takes that rule at its word and $^ expands to date.o ...
There are several ways to solve this problem. I'd do something like this:
OBJS := date.o jquery.autocomplete.o jquery.bullseye.o ...
OBJS := $(addprefix $(build),$(OBJS))
$(dest)main.js: $(OBJS)
cat $^ > $#
# you might have to tinker with this rule a little
$(build)%.o: %.js
php $(closure) $< $#
I have a "lib" directory in my applications main directory, which contains an arbitrary number of subdirectories, each having its own Makefile.
I would like to have a single Makefile in the main directory, that calls each subdirectory's Makefile. I know this is possible if I manually list the subdirs, but I would like to have it done automatically.
I was thinking of something like the following, but it obviously does not work. Note that I also have clean, test, etc. targets, so % is probably not a good idea at all.
LIBS=lib/*
all: $(LIBS)
%:
(cd $#; $(MAKE))
Any help is appreciated!
The following will work with GNU make:
LIBS=$(wildcard lib/*)
all: $(LIBS)
.PHONY: force
$(LIBS): force
cd $# && pwd
If there might be something other than directories in lib, you could alternatively use:
LIBS=$(shell find lib -type d)
To address the multiple targets issue, you can build special targets for each directory, then strip off the prefix for the sub-build:
LIBS=$(wildcard lib/*)
clean_LIBS=$(addprefix clean_,$(LIBS))
all: $(LIBS)
clean: $(clean_LIBS)
.PHONY: force
$(LIBS): force
echo make -C $#
$(clean_LIBS): force
echo make -C $(patsubst clean_%,%,$#) clean
There is also a way of listing sub-directories with gmake commands only, without using any shell commands:
test:
#echo $(filter %/, $(wildcard lib/*/))
This will list all sub-directories with trailing '/'. To remove it you can use the substitute pattern:
subdirs = $(filter %/, $(wildcard lib/*/))
test:
#echo $(subdirs:%/=%)
Then to actually create rules executing makefiles in each sub-directory you can use a small trick - a phony target in a non-existent directory. I think in this case an example will tell more than any explanation:
FULL_DIRS =$(filter %/, $(wildcard lib/*/))
LIB_DIRS =$(FULL_DIRS:%/=%)
DIRS_CMD =$(foreach subdir, $(LIB_DIRS), make-rule/$(subdir))
make-rule/%:
cd $* && $(MAKE)
all: DIRS_CMD
Basically, target 'all' lists all sub-directories as prerequisites. For example, if LIB_DIRS contained lib/folder1 lib/folder2 then the expansion would look like this:
all: make-rule/lib/folder1 make-rule/lib/folder2
Then 'make', in order to execute rule 'all', tries to match each prerequisite with an existing target. In this case the target is 'make-rule/%:', which uses '$*' to extract the string after 'make-rule/' and uses it as argument in the recipe. For example, the first prerequisite would be matched and expanded like this:
make-rule/lib/folder1:
cd lib/folder1 && $(MAKE)
What if you want to call different targets than all in an unknown number of subdirectories?
The following Makefile uses macros so create a forwarding dummy-target for a number of subdirectories to apply the given target from the command line to each of them:
# all direct directories of this dir. uses "-printf" to get rid of the "./"
DIRS=$(shell find . -maxdepth 1 -mindepth 1 -type d -not -name ".*" -printf '%P\n')
# "all" target is there by default, same logic as via the macro
all: $(DIRS)
$(DIRS):
$(MAKE) -C $#
.PHONY: $(DIRS)
# if explcit targets where given: use them in the macro down below. each target will be delivered to each subdirectory contained in $(DIRS).
EXTRA_TARGETS=$(MAKECMDGOALS)
define RECURSIVE_MAKE_WITH_TARGET
# create new variable, with the name of the target as prefix. it holds all
# subdirectories with the target as suffix
$(1)_DIRS=$$(addprefix $(1)_,$$(DIRS))
# create new target with the variable holding all the subdirectories+suffix as
# prerequisite
$(1): $$($1_DIRS)
# use list to create target to fullfill prerequisite. the rule is to call
# recursive make into the subdir with the target
$$($(1)_DIRS):
$$(MAKE) -C $$(patsubst $(1)_%,%,$$#) $(1)
# and make all targets .PHONY
.PHONY: $$($(1)_DIRS)
endef
# evaluate the macro for all given list of targets
$(foreach t,$(EXTRA_TARGETS),$(eval $(call RECURSIVE_MAKE_WITH_TARGET,$(t))))
Hope this helps. Really helpfull when dealing with paralelism: make -j12 clean all in a tree with makefiles having these targets... As always: playing with make is dangerous, different meta-levels of programming are too close together ,-)
I want to create directories using makefile. My project directory is like this
+--Project
+--output
+--source
+Testfile.cpp
+Makefile
I want to put all the objects and output into the respective output folder. I want to create folder structure which would be like this after compiling.
+--Project
+--output
+--debug (or release)
+--objs
+Testfile.o
+Testfile (my executable file)
+--source
+Testfile.cpp
+Makefile
I tried with several options, but could not succeed. Please help me to make directories using make file. I'm posting my Makefile for your consideration.
#---------------------------------------------------------------------
# Input dirs, names, files
#---------------------------------------------------------------------
OUTPUT_ROOT := output/
TITLE_NAME := TestProj
ifdef DEBUG
TITLE_NAME += _DEBUG
else
ifdef RELEASE
TITLE_NAME += _RELEASE
endif
endif
# Include all the source files here with the directory tree
SOURCES := \
source/TestFile.cpp \
#---------------------------------------------------------------------
# configs
#---------------------------------------------------------------------
ifdef DEBUG
OUT_DIR := $(OUTPUT_ROOT)debug
CC_FLAGS := -c -Wall
else
ifdef RELEASE
OUT_DIR := $(OUTPUT_ROOT)release
CC_FLAGS := -c -Wall
else
$(error no build type defined)
endif
endif
# Put objects in the output directory.
OUT_O_DIR := $(OUT_DIR)/objs
#---------------------------------------------------------------------
# settings
#---------------------------------------------------------------------
OBJS = $(SOURCES:.cpp=.o)
DIRS = $(subst /,/,$(sort $(dir $(OBJS))))
DIR_TARGET = $(OUT_DIR)
OUTPUT_TARGET = $(OUT_DIR)/$(TITLE_NAME)
CC_FLAGS +=
LCF_FLAGS :=
LD_FLAGS :=
#---------------------------------------------------------------------
# executables
#---------------------------------------------------------------------
MD := mkdir
RM := rm
CC := g++
#---------------------------------------------------------------------
# rules
#---------------------------------------------------------------------
.PHONY: all clean title
all: title
clean:
$(RM) -rf $(OUT_DIR)
$(DIR_TARGET):
$(MD) -p $(DIRS)
.cpp.o:
#$(CC) -c $< -o $#
$(OBJS): $(OUT_O_DIR)/%.o: %.cpp
#$(CC) -c $< -o $#
title: $(DIR_TARGET) $(OBJS)
In my opinion, directories should not be considered targets of your makefile, either in technical or in design sense. You should create files and if a file creation needs a new directory then quietly create the directory within the rule for the relevant file.
If you're targeting a usual or "patterned" file, just use make's internal variable $(#D), that means "the directory the current target resides in" (cmp. with $# for the target). For example,
$(OUT_O_DIR)/%.o: %.cpp
#mkdir -p $(#D)
#$(CC) -c $< -o $#
title: $(OBJS)
Then, you're effectively doing the same: create directories for all $(OBJS), but you'll do it in a less complicated way.
The same policy (files are targets, directories never are) is used in various applications. For example, git revision control system doesn't store directories.
Note: If you're going to use it, it might be useful to introduce a convenience variable and utilize make's expansion rules.
dir_guard=#mkdir -p $(#D)
$(OUT_O_DIR)/%.o: %.cpp
$(dir_guard)
#$(CC) -c $< -o $#
$(OUT_O_DIR_DEBUG)/%.o: %.cpp
$(dir_guard)
#$(CC) -g -c $< -o $#
title: $(OBJS)
This would do it - assuming a Unix-like environment.
MKDIR_P = mkdir -p
.PHONY: directories
all: directories program
directories: ${OUT_DIR}
${OUT_DIR}:
${MKDIR_P} ${OUT_DIR}
This would have to be run in the top-level directory - or the definition of ${OUT_DIR} would have to be correct relative to where it is run. Of course, if you follow the edicts of Peter Miller's "Recursive Make Considered Harmful" paper, then you'll be running make in the top-level directory anyway.
I'm playing with this (RMCH) at the moment. It needed a bit of adaptation to the suite of software that I am using as a test ground. The suite has a dozen separate programs built with source spread across 15 directories, some of it shared. But with a bit of care, it can be done. OTOH, it might not be appropriate for a newbie.
As noted in the comments, listing the 'mkdir' command as the action for 'directories' is wrong. As also noted in the comments, there are other ways to fix the 'do not know how to make output/debug' error that results. One is to remove the dependency on the the 'directories' line. This works because 'mkdir -p' does not generate errors if all the directories it is asked to create already exist. The other is the mechanism shown, which will only attempt to create the directory if it does not exist. The 'as amended' version is what I had in mind last night - but both techniques work (and both have problems if output/debug exists but is a file rather than a directory).
Or, KISS.
DIRS=build build/bins
...
$(shell mkdir -p $(DIRS))
This will create all the directories after the Makefile is parsed.
make in, and off itself, handles directory targets just the same as file targets. So, it's easy to write rules like this:
outDir/someTarget: Makefile outDir
touch outDir/someTarget
outDir:
mkdir -p outDir
The only problem with that is, that the directories timestamp depends on what is done to the files inside. For the rules above, this leads to the following result:
$ make
mkdir -p outDir
touch outDir/someTarget
$ make
touch outDir/someTarget
$ make
touch outDir/someTarget
$ make
touch outDir/someTarget
This is most definitely not what you want. Whenever you touch the file, you also touch the directory. And since the file depends on the directory, the file consequently appears to be out of date, forcing it to be rebuilt.
However, you can easily break this loop by telling make to ignore the timestamp of the directory. This is done by declaring the directory as an order-only prerequsite:
# The pipe symbol tells make that the following prerequisites are order-only
# |
# v
outDir/someTarget: Makefile | outDir
touch outDir/someTarget
outDir:
mkdir -p outDir
This correctly yields:
$ make
mkdir -p outDir
touch outDir/someTarget
$ make
make: 'outDir/someTarget' is up to date.
TL;DR:
Write a rule to create the directory:
$(OUT_DIR):
mkdir -p $(OUT_DIR)
And have the targets for the stuff inside depend on the directory order-only:
$(OUT_DIR)/someTarget: ... | $(OUT_DIR)
All solutions including the accepted one have some issues as stated in their respective comments. The accepted answer by #jonathan-leffler is already quite good but does not take into effect that prerequisites are not necessarily to be built in order (during make -j for example). However simply moving the directories prerequisite from all to program provokes rebuilds on every run AFAICT.
The following solution does not have that problem and AFAICS works as intended.
MKDIR_P := mkdir -p
OUT_DIR := build
.PHONY: directories all clean
all: $(OUT_DIR)/program
directories: $(OUT_DIR)
$(OUT_DIR):
${MKDIR_P} $(OUT_DIR)
$(OUT_DIR)/program: | directories
touch $(OUT_DIR)/program
clean:
rm -rf $(OUT_DIR)
I've just come up with a fairly reasonable solution that lets you define the files to build and have directories be automatically created. First, define a variable ALL_TARGET_FILES that holds the file name of every file that your makefile will be build. Then use the following code:
define depend_on_dir
$(1): | $(dir $(1))
ifndef $(dir $(1))_DIRECTORY_RULE_IS_DEFINED
$(dir $(1)):
mkdir -p $$#
$(dir $(1))_DIRECTORY_RULE_IS_DEFINED := 1
endif
endef
$(foreach file,$(ALL_TARGET_FILES),$(eval $(call depend_on_dir,$(file))))
Here's how it works. I define a function depend_on_dir which takes a file name and generates a rule that makes the file depend on the directory that contains it and then defines a rule to create that directory if necessary. Then I use foreach to call this function on each file name and eval the result.
Note that you'll need a version of GNU make that supports eval, which I think is versions 3.81 and higher.
given that you're a newbie, I'd say don't try to do this yet. it's definitely possible, but will needlessly complicate your Makefile. stick to the simple ways until you're more comfortable with make.
that said, one way to build in a directory different from the source directory is VPATH; i prefer pattern rules
OS independence is critical for me, so mkdir -p is not an option. I created this series of functions that use eval to create directory targets with the prerequisite on the parent directory. This has the benefit that make -j 2 will work without issue since the dependencies are correctly determined.
# convenience function for getting parent directory, will eventually return ./
# $(call get_parent_dir,somewhere/on/earth/) -> somewhere/on/
get_parent_dir=$(dir $(patsubst %/,%,$1))
# function to create directory targets.
# All directories have order-only-prerequisites on their parent directories
# https://www.gnu.org/software/make/manual/html_node/Prerequisite-Types.html#Prerequisite-Types
TARGET_DIRS:=
define make_dirs_recursively
TARGET_DIRS+=$1
$1: | $(if $(subst ./,,$(call get_parent_dir,$1)),$(call get_parent_dir,$1))
mkdir $1
endef
# function to recursively get all directories
# $(call get_all_dirs,things/and/places/) -> things/ things/and/ things/and/places/
# $(call get_all_dirs,things/and/places) -> things/ things/and/
get_all_dirs=$(if $(subst ./,,$(dir $1)),$(call get_all_dirs,$(call get_parent_dir,$1)) $1)
# function to turn all targets into directories
# $(call get_all_target_dirs,obj/a.o obj/three/b.o) -> obj/ obj/three/
get_all_target_dirs=$(sort $(foreach target,$1,$(call get_all_dirs,$(dir $(target)))))
# create target dirs
create_dirs=$(foreach dirname,$(call get_all_target_dirs,$1),$(eval $(call make_dirs_recursively,$(dirname))))
TARGETS := w/h/a/t/e/v/e/r/things.dat w/h/a/t/things.dat
all: $(TARGETS)
# this must be placed after your .DEFAULT_GOAL, or you can manually state what it is
# https://www.gnu.org/software/make/manual/html_node/Special-Variables.html
$(call create_dirs,$(TARGETS))
# $(TARGET_DIRS) needs to be an order-only-prerequisite
w/h/a/t/e/v/e/r/things.dat: w/h/a/t/things.dat | $(TARGET_DIRS)
echo whatever happens > $#
w/h/a/t/things.dat: | $(TARGET_DIRS)
echo whatever happens > $#
For example, running the above will create:
$ make
mkdir w/
mkdir w/h/
mkdir w/h/a/
mkdir w/h/a/t/
mkdir w/h/a/t/e/
mkdir w/h/a/t/e/v/
mkdir w/h/a/t/e/v/e/
mkdir w/h/a/t/e/v/e/r/
echo whatever happens > w/h/a/t/things.dat
echo whatever happens > w/h/a/t/e/v/e/r/things.dat
See https://www.oreilly.com/library/view/managing-projects-with/0596006101/ch12.html
REQUIRED_DIRS = ...
_MKDIRS := $(shell for d in $(REQUIRED_DIRS); \
do \
[[ -d $$d ]] || mkdir -p $$d; \
done)
$(objects) : $(sources)
As I use Ubuntu, I also needed add this at the top of my Makefile:
SHELL := /bin/bash # Use bash syntax
I use the makefiles in windows environment and my simple solution is as follows,
Create a target makedir and add it as a prerequisites to where ever it is required.
# Default goal
all: gccversion makedir build finalize list sizeafter completed
The makedir target is (applicable only in windows environment)
makedir:
#IF NOT EXIST $(subst /,\,$(BUILD_DIR)) mkdir $(subst /,\,$(BUILD_DIR)) 2> NULL
#IF NOT EXIST $(subst /,\,$(OUTPUT_DIR)) mkdir $(subst /,\,$(OUTPUT_DIR)) 2> NULL
#IF NOT EXIST $(subst /,\,$(DEP_DIR)) mkdir $(subst /,\,$(DEP_DIR)) 2> NUL
#IF NOT EXIST $(subst /,\,$(OBJ_DIR)) mkdir $(subst /,\,$(OBJ_DIR)) 2> NUL
$(subst /,\,$(BUILD_DIR)) converts the directory separator / to \ and
mkdir $(subst /,\,$(BUILD_DIR)) 2> NUL redirects the error if any.
src_dir := src
obj_dir := obj
build_dir := build
dirs := $(src_dir) $(obj_dir) $(build_dir) # new variable
all: $(dirs) $(other_dependencies) # added dependency (*before* any others)
$(dirs): # rule which makes missing directories
mkdir $#
Won't clutter your terminal with "cannot create directory" error messages. If the directories exist, they don't need to be built.
Works like any other dependency, only requires one rule and one variable.