There is no single correct answer, nor are there any strong conventions assumed or enforced by any Go tooling at this time.
Typically I start by assuming that the files I need are located in the same directory from where the program will be run. For instance, suppose I need conf.json
for myprog.go
; then both of those files live together in the same directory and it works to just run something like
go build -o myprog && ./myprog
When I deploy the code, the myprog
binary and conf.json
live together on the server. The run/supervisor script needs to cd
to that directory and then run the program.
This also works when you have a lot of resources; for instance, if you have a webserver with JS, CSS, and images, you just assume they're relative to cwd in the code and deploy the resource directories along with the server binary.
Another alternative to assuming a cwd is to have a -conf
flag which the user can use to specify a configuration file. I typically use this for distributing command-line tools and open-source server applications that require a single configuration file. You could even use an -assets
flag or something to point to a whole tree of resource files, if you wanted.
Finally, one more approach is to not have any resource files. go-bindata is a useful tool that I've used for this purpose -- it just encodes some data as bytes into a Go source file. That way it's all baked into your binary. I think this method is most useful when the resource data will rarely or never change, and is pretty small. (Otherwise you're going to be shipping around huge binaries.) One (kind of silly) example of when I've used go-bindata in the past was for baking a favicon into a really simple server which didn't otherwise require any extra files besides the server binary.