Need some help packaging parts of a scientific workflow

Need some help packaging parts of a scientific workflow

[Kyle Andrews]

Dear Guix,

Part of my scientific workflow involves compiling a small Racket script
for a command line program into its executable and placing that on
PATH. I had bundled this script inside an R package which made sure it
got compiled and everything was correctly configured at library load
time.

From hanging around here for a bit, I have learned this practice is
probably a bad thing for reproducibility.  I should separate my support
programs out from my R package rather than leaving them bundled in.

From reading the documentation a lot, I think the actual compilation
step can be done using the "invoke" procedure like so:

```
(invoke "raco" "exe" "{package_location}/custom-shell-tool.rkt")
```

What I'm struggling with the most is understanding all the boilerplate
code I need to place around that fundamental call.

Below is a text-painting of the state of my misunderstanding:

```
(package
 (name "custom-shell-tool")
 (version "1")
 (source
  (local-file "package_location")) ; how to refer to local files?
 (build-system copy-build-system)
 (arguments
  (list
   #:phases
   #~(modify-phases %standard-phases
      (add-before 'install 'compile 
       (lambda _
        (invoke "raco" "exe"
         (string-append
          #$package-folder ; how to refer to the build itself?
          "custom-shell-tool.rkt"))))))
   #:install-plan
   #~'(("custom-shell-tool" "bin/")))
  (home-page #f)
  (synopsis
   "A custom shell tool needed only for the niche workflows I write")
  (description "This package is different from the ones in Guix.")
  (license #f))
```

Would you please help me translate this text painting into a working
package?

I'm especially interested in figuring out how I can productively learn
to experiment productively with this stuff for myself.

Thanks for your help,
Kyle

Re: Need some help packaging parts of a scientific workflow

[Philip McGrath]

Hi Kyle,

On Wed, Mar 22, 2023, at 5:29 PM, Kyle Andrews wrote:
> Dear Guix,
>
> Part of my scientific workflow involves compiling a small Racket script
> for a command line program into its executable and placing that on
> PATH.

I am always glad to hear of more people using Guix and Racket together!

> I had bundled this script inside an R package which made sure it
> got compiled and everything was correctly configured at library load
> time.
>

Tangential to your actual question, I think this is not necessarily a terrible practice. There is not much difference between running `my-script` and running `racket -y "path/to/my-script.rkt"`, and, if you do that or `raco make` during the build process of your R package, you'll get compiled files properly. There are tradeoffs to weigh, including support for R users without Guix. But there are also good reasons to decide to separate the Racket script from the R library, so that's what I'll explain below.

> From reading the documentation a lot, I think the actual compilation
> step can be done using the "invoke" procedure like so:
>
> ```
> (invoke "raco" "exe" "{package_location}/custom-shell-tool.rkt")
> ```
>
> What I'm struggling with the most is understanding all the boilerplate
> code I need to place around that fundamental call.
>

I'll start with a working example suitable for `guix build -f`, then answer your specific questions.

```
;; SPDX-License-Identifier: (CC0-1.0 OR (Apache-2.0 OR MIT))
;; SPDX-FileCopyrightText: Philip McGrath <philip@philipmcgrath.com>

(use-modules
 (gnu packages racket)
 (guix build-system copy)
 (guix gexp)
 (guix packages))

(package
  (name "racket-hello")
  (version "1.0")
  (source (plain-file "hello.rkt"
                      "#lang racket (displayln '|Hello from Racket!|)"))
  (inputs (list racket))
  (build-system copy-build-system)
  (arguments
   (list
    #:install-plan #~'(("hello" "bin/"))
    #:phases
    #~(modify-phases %standard-phases
        (add-before 'install 'build
          (lambda args
            (invoke "raco" "exe" "hello.rkt"))))))
  (home-page #f)
  (synopsis "Hello world in Racket")
  (description
   "This is a trivial example of using @code{raco exe} with Guix.")
  (license #f))
```

In fact this package would be a reasonable candidate for `trivial-build-system`, but I've stuck with `copy-build-system` because the boilerplate for `trivial-build-system` is very different than for all other build systems.

Likewise, I'm assuming you know how you want to build your executable, but you might consider the `--launcher` flag for `raco exe` and an explicit call to `raco make`: in particular, it might take use less total disk space than an ELF executable.

Note that you do have to use `racket`, not `racket-minimal`, because `racket-minimal` doesn't include the `raco exe` command.

>  (source
>   (local-file "package_location")) ; how to refer to local files?

In general, `local-file` is the right mechanism; specifics depend on your situation, including how you are expecting your package definition to be used. For a single file, `(local-file "path/to/script.rkt")` is probably what you want, where the path is relative to the Guile file containing the `local-file` expression. For a directory, consider the `#:recursive?` and `#:select?` arguments.

>         (invoke "raco" "exe"
>          (string-append
>           #$package-folder ; how to refer to the build itself?
>           "custom-shell-tool.rkt"))))))

The `unpack` phase from `gnu-build-system` handles this: if the package source is a directory, you're inside a copy of it; if it's a file, you're in a temporary directory containing it.

>
> I'm especially interested in figuring out how I can productively learn
> to experiment productively with this stuff for myself.
>

Personally, I often insert a phase that calls `error` to stop the build, perhaps printing out interesting values first, and use `guix build --keep-failed` to explore the build environment.

-Philip

Re: Need some help packaging parts of a scientific workflow

[Kyle]

Thanks, Philip! Your descriptions really made it click for me.

Might you give a small example of what you mean by adding an error phase? That sounds like an important debugging strategy, but I can't quite grasp what it would actually look like. I'm curious what kinds of data you are printing out. 

On March 23, 2023 5:30:11 PM EDT, Philip McGrath <philip@philipmcgrath.com> wrote:
>Hi Kyle,
>
>On Wed, Mar 22, 2023, at 5:29 PM, Kyle Andrews wrote:
>> Dear Guix,
>>
>> Part of my scientific workflow involves compiling a small Racket script
>> for a command line program into its executable and placing that on
>> PATH.
>
>I am always glad to hear of more people using Guix and Racket together!
>
>> I had bundled this script inside an R package which made sure it
>> got compiled and everything was correctly configured at library load
>> time.
>>
>
>Tangential to your actual question, I think this is not necessarily a terrible practice. There is not much difference between running `my-script` and running `racket -y "path/to/my-script.rkt"`, and, if you do that or `raco make` during the build process of your R package, you'll get compiled files properly. There are tradeoffs to weigh, including support for R users without Guix. But there are also good reasons to decide to separate the Racket script from the R library, so that's what I'll explain below.
>
>> From reading the documentation a lot, I think the actual compilation
>> step can be done using the "invoke" procedure like so:
>>
>> ```
>> (invoke "raco" "exe" "{package_location}/custom-shell-tool.rkt")
>> ```
>>
>> What I'm struggling with the most is understanding all the boilerplate
>> code I need to place around that fundamental call.
>>
>
>I'll start with a working example suitable for `guix build -f`, then answer your specific questions.
>
>```
>;; SPDX-License-Identifier: (CC0-1.0 OR (Apache-2.0 OR MIT))
>;; SPDX-FileCopyrightText: Philip McGrath <philip@philipmcgrath.com>
>
>(use-modules
> (gnu packages racket)
> (guix build-system copy)
> (guix gexp)
> (guix packages))
>
>(package
>  (name "racket-hello")
>  (version "1.0")
>  (source (plain-file "hello.rkt"
>                      "#lang racket (displayln '|Hello from Racket!|)"))
>  (inputs (list racket))
>  (build-system copy-build-system)
>  (arguments
>   (list
>    #:install-plan #~'(("hello" "bin/"))
>    #:phases
>    #~(modify-phases %standard-phases
>        (add-before 'install 'build
>          (lambda args
>            (invoke "raco" "exe" "hello.rkt"))))))
>  (home-page #f)
>  (synopsis "Hello world in Racket")
>  (description
>   "This is a trivial example of using @code{raco exe} with Guix.")
>  (license #f))
>```
>
>In fact this package would be a reasonable candidate for `trivial-build-system`, but I've stuck with `copy-build-system` because the boilerplate for `trivial-build-system` is very different than for all other build systems.
>
>Likewise, I'm assuming you know how you want to build your executable, but you might consider the `--launcher` flag for `raco exe` and an explicit call to `raco make`: in particular, it might take use less total disk space than an ELF executable.
>
>Note that you do have to use `racket`, not `racket-minimal`, because `racket-minimal` doesn't include the `raco exe` command.
>
>>  (source
>>   (local-file "package_location")) ; how to refer to local files?
>
>In general, `local-file` is the right mechanism; specifics depend on your situation, including how you are expecting your package definition to be used. For a single file, `(local-file "path/to/script.rkt")` is probably what you want, where the path is relative to the Guile file containing the `local-file` expression. For a directory, consider the `#:recursive?` and `#:select?` arguments.
>
>>         (invoke "raco" "exe"
>>          (string-append
>>           #$package-folder ; how to refer to the build itself?
>>           "custom-shell-tool.rkt"))))))
>
>The `unpack` phase from `gnu-build-system` handles this: if the package source is a directory, you're inside a copy of it; if it's a file, you're in a temporary directory containing it.
>
>>
>> I'm especially interested in figuring out how I can productively learn
>> to experiment productively with this stuff for myself.
>>
>
>Personally, I often insert a phase that calls `error` to stop the build, perhaps printing out interesting values first, and use `guix build --keep-failed` to explore the build environment.
>
>-Philip