Running make produces hard to decipher error messages

[yuvipanda]

The title could also be because I'm generally scared of make :)

From the master branch, if I type make I get the following error:

$ make
jq: error: Could not open file : No such file or directory
[+] Building 0.0s (0/0)                                                                                           
error: open : no such file or directory
jq: error: Could not open file : No such file or directory
jq: error: Could not open file : No such file or directory
make: Nothing to be done for `all'.

I'm not sure what is wrong nor how exactly to interpret this.

[eitsupi]

I don't think this is a make's failure, just that the default target all is a dummy target where nothing actually happens.

rocker-versioned2/Makefile

Lines 5 to 6 in 485c8bb

	all:

The jq error I have seen many times and am not sure where it is coming from.
Maybe it's because there are a few places where docker command are handled by jq.
Like this:

rocker-versioned2/Makefile

Line 39 in 485c8bb

pull-image-all: $(foreach I, $(shell jq '.target[].tags[]' -r $(BAKE_JSON) | sed -e 's/:/\\:/g'), pull-image/$(I))

[yuvipanda]

Yeah, definitely not an issue with make itself - I have updated the title to be clearer that it's running make that provides this output.

make setup also does the same.

And yeah, I too am not sure where it's coming from.

[yuvipanda]

I'm also not sure what make target to use to actually locally build all containers.

[eitsupi]

In fact, most parts of this Makefile are intended to be used on CI and are not needed locally.

I'm also not sure what make target to use to actually locally build all containers.

The point here is that r-ver is the only Dockerfile that is copied files into the container.

rocker-versioned2/dockerfiles/r-ver_devel.Dockerfile

Line 19 in 485c8bb

COPY scripts /rocker_scripts

This means that in order to build other images, we must first build r-ver and bring the updated rocker_scripts into the container.

In practice, this is too much trouble, so we need to write our own Dockerfile that copies the rocker_scripts into the container, overwrites the rocker_scripts in the container, and builds it.

Like this:

rocker-versioned2/tests/rocker_scripts/Dockerfile

Lines 1 to 10 in 485c8bb

	ARG base_image="rocker/r-ver"
	ARG tag="latest"
	FROM ${base_image}:${tag}
	COPY tests/rocker_scripts/test.sh /test.sh
	COPY scripts /rocker_scripts
	ARG script_name=install_rstudio.sh
	ARG script_arg=skip
	RUN /test.sh ${script_name} ${script_arg}

[eitsupi]

Honestly, I'm tired of using Make for task runners like this, and this Makefile isn't very good since I wrote it when I knew very little about Make...

I would definitely use Task instead of Make now.

[yuvipanda]

I'm personally a very large proponent of not using Make in general :D

[eddelbuettel]

Hm, no, make is rock solid. An ACM Software Systems Award (in 2003, just like S / R in 1998) is rarely wrong. May I point out (cf #719) that this is a fairly complex project and that using JSON (confession time: I actually think that is not a great choice for config files, apparently I am near alone in that view) in all its glory and dependencies may be related ? Anyway.

Famous quip about programming languages (by Stroustrup if memory serves) is that there two kinds: the ones people complain about, and the ones nobody uses. Same for build systems. We all have cmake too but whatyagottado ....

[yuvipanda]

@eddelbuettel would love some help with the error message!

[eddelbuettel]

@yuvipanda I hear you but we have a large JSON shaped gap between the two of us! Those are jq errors so ball in your court (or in @eitsupi for the wide selection of R packages). make runs subshells, I find one can generally substitute direct commands. Would love to help, can't here by toolkit choice.

[yuvipanda]

Would love to help, can't here by toolkit choice.

That makes the two of us, from different sides :)