different output in Makefile rule - makefile

In the following Makefile
f = echo $(1)
t:
$(call f,"a \
b"); \
$(call f,"a \
b")
there is only one TAB, at the beginning of line 4 (the first call f)
no blanks at the end of the lines
The output is
a b
a b
The two calls are identical (except for the initial TAB). Why the output are different ?

I suspect the answer to why the output is different is that this is a small bug in GNU Make's backslash-newline whitespace collapsing. The relevant code is in new_job():
/* Discard any preceding whitespace that has
already been written to the output. */
while (out > ref
&& isblank ((unsigned char)out[-1]))
--out;
I don't know this code, but the in and ref pointers operate in lockstep, so comparing out to ref seems wrong. Replacing out > ref with out > cmds->command_lines[i] corrects your problem and might not be a completely incorrect change to make to the code.
Whether this is a GNU Make bug or not, it's not really a great idea for your makefile to depend so intimately on the vagaries of whitespace removal. If the whitespace is truly significant in your recipe, you could code this defensively as
f = echo $(strip $(1))
and avoid any problems at a stroke.

Related

Use shell arguments inside GNU Make function

Is is possible to use shell arguments inside a makefile function?
e.g.
mywords:=hello there neighbour
myrecipe:
for ((i = 0 ; i < 3 ; i++)); do \
this_word=$(word $${i}, $(mywords)); \
done
Except myrecipe gives the error non-numeric first argument to 'word' function: '${i}'. Stop
Remember that make first expands the entire recipe (yes, that's all of the recipe lines) before ever doing anything with the shell.
Only after this full expansion does make then look for individual recipe lines, running each one in turn.
Thus make behaves sensibly when you have a macro that expands to more than one line.
It also hints that you should use make functionality wherever possible.
mywords := hello there neighbour
define mkcommand
echo $1
something $1
endef
myrecipe:
$(foreach _,${mywords},$(call mkcommand,$_))
Now, when you ask make to build myrecipe,
make sees something like:
myrecipe:
echo hello
something hello
echo there
something there
⋮
IMHO it's nearly always a mistake if you find yourself writing shell loops in a makefile.

How to know if a makefile variable is a string of char or numbers?

It probably sounds very elementary but I am unable to find a way to classify a makefile variable into text or number. My pseudocode is like this:
ifeq ($N, 'numeric')
CFLAGS+=-D$N
endif
How to do this? I am using the GNU Make (in cygwin/Windows). I read the make.pdf that comes with it but could not find a way.
Thanks in Advance
EDIT: adopted a suggestion by bobbogo that does not depend on the number of characters to purge.
I assume you use GNU make. Here is a make-only solution, without calling the shell. For performance reasons, depending on your use of it, it can be preferable. Moreover, it does not depend on which shell make uses. Last but not least, it uses recursion and I like recursion:
define PURGE
$(if $(2),$(call PURGE,$(subst $(firstword $(2)),,$(1)),$(filter-out $(firstword $(2)),$(2))),$(1))
endef
DIGITS := 0 1 2 3 4 5 6 7 8 9
define IS_NOT_A_NUMBER
$(call PURGE,$(1),$(DIGITS))
endef
CFLAGS += $(if $(call IS_NOT_A_NUMBER,$(N)),,-D$(N))
all:
$(info N=$(N) => CFLAGS=$(CFLAGS))
Demo:
host> make N=12345
N=12345 => CFLAGS=-D12345
make: 'all' is up to date.
host> make N=foobar
N=foobar => CFLAGS=
make: 'all' is up to date.
Explanation: PURGE is a recursive macro that takes two arguments. The first one ($(1)) is a string to test, the second one ($(2)) is a list of words to match. If $(2) is the empty list PURGE returns $(1). Else, it calls itself with two new parameters:
the value of $(1) where the first word of $(2) has been substituted by nothing,
$(2) from which the first word has been removed
and returns the result. So, if you call PURGE with a string and the list of all digits, it returns the empty string if and only if the string contained only digits.
All make variables are strings. To find out whether a string is in fact a number, you need some elementary text analysis functions. GNU make itself does not offer anything convenient in this area, but you could run a shell command to do the job, perhaps like this:
define is_number
$(shell test '$(1)' -eq '$(1)' 2>/dev/null && echo yes || echo no)
endef
ifeq ($(call is_number, $(N)),yes)
default:
#echo N is a number
else
default:
#echo N is not a number
endif
This results in:
$ make N=5
N is a number
$ make N=string
N is not a number
However, such string processing can be quite unreliable if the string contains special characters.

Recursive make: correct way to insert `$(MAKEFLAGS)`

How can I use $(MAKEFLAGS) (or another way of passing variables defined on the command line to sub-make) in a way that supports invocation from shell with both make VAR=val and make -args?
I need my subprojects configurable, but I hate autotools, so I'm using make variables for this, e.g. invoking from shell:
$ make USE_SSE3=1
and USE_SSE3 needs to apply to all builds in all sub-makefiles.
The manual states that:
if you do ‘make -ks’ then MAKEFLAGS gets the value ‘ks’.
Therefore I'm using -$(MAKEFLAGS) (with a dash prefix) in my Makefile.
However, that expands into invalid arguments when variables with no flags are used. If I run:
$ make FOO=bar
then sub-make gets invalid -FOO=bar. OTOH without the dash prefix variable definitions work, then but make -s, etc. don't.
Is there a syntax/variable/hack that makes passing of arguments and lone variable definitions work with sub-makefiles?
The legacy $(MKFLAGS) doesn't have the weird dash prefix problem, but it doesn't include variable definitions either. I've tried fixing the variable with $(patsubst), but that only made things worse by trimming whitespace.
I need the solution to be compatible with the outdated GNU Make 3.81 shipped with Mac OS X Mavericks.
foo:
$(MAKE) -C subproject -$(MAKEFLAGS)
$ make foo -s # MAKEFLAGS = 's'
$ make foo BAR=baz # MAKEFLAGS = 'BAR=baz'
$ make foo -j8 # MAKEFLAGS = ' --job-server=…'
You shouldn't set MAKEFLAGS at all. Why do you want to? You didn't give any reason to do so.
MAKEFLAGS is intended, really, to be an internal implementation passing arguments from a parent make to a child make. It's not intended, generally, to be modified by a makefile. About the only thing you can usefully do to it is add new flags.
If you just run the recursive make using the $(MAKE) variable rather than hardcoding make, it will Just Work:
recurse:
#$(MAKE) all FOO=bar
or whatever.
Years too late I got your answer if I got it right.
You can construct $(MAKEARGS) manually yourself like:
MAKEARGS := $(strip \
$(foreach v,$(.VARIABLES),\
$(if $(filter command\ line,$(origin $(v))),\
$(v)=$(value $(v)) ,)))
MAKEARGS := assign static
strip cleans leading and trailing whitespaces.
foreach v iterate over all variable names.
origin $(v) check if variable origin is "command line".
$(v)=$(value $(v)) output env assignment string.
Alternatively you can unpick the $(MAKEFLAGS) like:
MAKEARGS := $(wordlist 2,$(words $(MAKEFLAGS)),$(MAKEFLAGS))
MAKEFLAGS := $(firstword $(MAKEFLAGS))
Which can leave you with cleaner code for further recursions IMHO. I say this because I sometimes need to keep apart arguments and flags in certain cases. Especially as you get caught in debugging a recursion djungle.
But for any specific case one should consult the manual about recursive options processing.
Changing the $(MAKEFLAGS) can lead to unwanted malfunction.
Another useful information for the willing user could be that the $(MAKEFLAGS) variable is basically the whole argument list passed to make, not only the flag characters. So $(info MAKEFLAGS = $(MAKEFLAGS)) can give you something like:
MAKEFLAGS = rRw -- VAR=val
Cheers
To check if -B is present in make flags i do :
BB_CLOBBER := $(shell echo $(MAKEFLAGS) | grep wB)
ifeq (,$(BB_CLOBBER))
# also force clobber make if these files are missing
BB_CLOBBER := $(shell (test -e $(bb_gen)/minimal/.config && test -e $(bb_gen)/full/.config) || echo "B")
endif
bb_prepare:
ifneq (,$(BB_CLOBBER))
#rm -rf $(bb_gen)/full
...

makefile: remove duplicate words without sorting

Is there a possibility to remove duplicates in a list of words without sorting in a makefile?
$(sort foo bar lose)
does remove duplicates (which is for me the main functionality in this case), but also sorts (for me an unfortunate side effect in this case). I want to avoid that.
[update]
bobbogo's answer works very nicely. Just remember to use define uniq for v3.81 and (did not check this) define uniq = for later versions.
larsmans' answer works very nicely too if your record separator is not a space, e.g. if you want to remove duplicates from _foo_bar_lose_lose_bar_baz_ or the like. Just remember to use the RS and ORS awk options instead of tr, and wrap it all with $(firstword $(shell ... ))
Boring $eval based method:
define uniq =
$(eval seen :=)
$(foreach _,$1,$(if $(filter $_,${seen}),,$(eval seen += $_)))
${seen}
endef
w := z z x x y c x
$(info $(sort $w))
$(info $(call uniq,$w))
Extremely fiendish make standard library recursive call (recursive make considered extremely fiendish?):
uniq = $(if $1,$(firstword $1) $(call uniq,$(filter-out $(firstword $1),$1)))
It's worth noting that no variables are damaged in this second formulation (see seen in the first). It is preferable just for that (given the lack of locals in make)!
EDIT
My obscure comment about recursive make above seems to have muddied the waters somewhat.
"Recursive" in the context of this post means recursive function.
It really has nothing to do with the execrable recursive make.
The latter (recursive) definition of uniq is extremely nice, performant, small, and is definitely the one to use.
Depends on where you need it and whether you use GNU make. If you just want to uniq the list of target prerequisites, it's as easy as (http://www.gnu.org/software/make/manual/make.html#Quick-Reference) :
The value of $^ omits duplicate prerequisites, while $+ retains them and preserves their order.
So, a rule like
exe: $(OBJS)
$(LD) -o $# $^
will filter duplicates from $(OBJS) automagically, while still leaving order of other items the same.
You could echo the words through awk:
echo foo bar foo baz bar | tr ' ' '\n' | awk '!a[$0]++'
Deduping one-liner taken from catonmat.
(Don't forget to double the $ to $$ in a Makefile.)
The following works for me under GNU make v3.82:
uniq = $(eval _uniq := $1)$(strip $(foreach _,$(_uniq),$(if $(filter $_,$(_uniq)),$(eval _uniq := $(filter-out $_,$(_uniq)))$_)))
It doesn't modify its input by creating a copy in _uniq, and it's not recursive.

functions in makefiles?

I have a slew of makefile targets that do the same thing:
${SOME_FILE}:
${FILES} | ${DIST_DIR}
##cat ${FILES} | \
sed 's/#DATE/'"${DATE}"'/' | \
sed 's/#VERSION/'"${CR_VER}"'/' \
> ${OUT_FILE};
where ${FILES} and ${OUT_FILE} are the only things changing. I'm trying to figure out if it's possible to simplify these targets to something like:
${SOME_FILE}:
compile(${FILES},${OUT_FILE})
Thanks for any insight.
GNU make has this:
http://www.gnu.org/software/make/manual/make.html#Call-Function
To define a multi-line function, you would use this syntax:
http://www.gnu.org/software/make/manual/make.html#Canned-Recipes
Links to docs (like in the accepted answer) are good but good example is better :)
define my_func
$(eval $#_PROTOCOL = "https:")
$(eval $#_HOSTNAME = $(1))
$(eval $#_PORT = $(2))
echo "${$#_PROTOCOL}//${$#_HOSTNAME}:${$#_PORT}/"
endef
my-target:
#$(call my_func,"example.com",8000)
Take into consideration the following:
There are no custom "functions" in Makefile language. So "my_func" is actually a variable that simply contains a text.
That "function" doesn't have its own scope. All the content of that "function" is copied into the recipe as is. All variables in the body will be used after that as a part of the recipe.
Don't use spaces near the commas to prettify param list of "call" function.
Args are passed like $(1), $(2), ... that is not very handy, so re-declare them with meaningful names. This is optional but recommended for bigger "function" bodies.
Variables are declared with $#_ prefix that makes them "local" to the rule (actually not local but prefixed by the target name).
So we have imitation of functions with imitation of local variables but generally this works good.
If you don't want to restrict yourself to GNUmake, your best bet is probably to generate makefile fragments yourself and then include them.

Resources