[PATCH] gnu: Add redo-apenwarr

Change-Id: Ied142a7dc3e9baf9babdeff046f350e647a7a5cc

---

gnu/packages/build-tools.scm | 110 +++++++++++++++++++++++++++++++++++

1 file changed, 110 insertions(+)

Hi Massimo,

Thank you for your submission. I am adding some specific notes about the package definition followed by some high-level notes about the overall package and the patch.

On 3/8/24 16:00, Massimo Zaniboni wrote:

`guix lint redo-apenwarr` reports an warning about the package name not being included in the source. This can be fixed by using the [file-name field in the origin](https://guix.gnu.org/manual/en/html_node/origin-Reference.html).This is helpful so that the store path clearly identifies the package it is related to. The README for this package mentions that there are other implementations of the redo build system. If an alternative implementation is added at some point in the future then store paths that simply identify themselves as "redo" will be ambiguous. Linting is one of the steps mentioned in the "[Submitting Patches](https://guix.gnu.org/manual/en/html_node/Submitting-Patches.html)" section of the manual. If the output from this tool or any of the other steps is unclear please let me know and I'll do my best to help!

Why is parallel building/testing disabled? Does this cause a build failure? If so, it would be preferable to find a way to fix it. If it cannot be fixed please add a comment explaining the problem.

Please make this modify-phases call into a [G-Expression](https://guix.gnu.org/manual/en/html_node/G_002dExpressions.html),as this is the updated convention and will become relevant to a later note. You will probably also want to change the arguments list from using a backtick to calling the `list` function.

My understanding is that the patch-source-shebangs phase, which runs before build in gnu-build-system, should patch this file. Do you know why it was necessary to add this line?

I see that git is included in native-inputs already, which should make it available in the build phase. If this is still a problem then I think there is something else we need to fix, rather than constructing the file manually.

In contrast to the patch-shebang call above, I think that this substitution is necessary because the shebang is in the contents of the script not the leading line (at least, this is the case for bin/default.do). However, the preferred way to locate binaries is with G-Expressions rather than looking them up dynamically. In this case I would expect the call to be `#$([file-append](https://guix.gnu.org/manual/en/html_node/G_002dExpressions.html#index-file_002dappend)bash "/bin/sh")`. There are some other places in the package where this also applies, but I will only annotate the ones that are meaningfully different.

I don't think the format call is necessary here?

While guix packages do not typically patch every single file to use absolute paths I think that it is better to use file-append since a substitution is happening here anyway.

As above, file-append would be appropriate here. --- The output of the build works well. I was able to run the ["Hello, world!" example from the manual](https://redo.readthedocs.io/en/latest/cookbook/hello/)from a [container](https://guix.gnu.org/manual/en/html_node/Invoking-guix-shell.html#index-container) with only redo-apenwall, coreutils, and gcc-toolchain (coreutils might not have been necessary, but it was convenient for me and reasonable for a minimal container IMO). Additionally, building with [`--rounds=2`](https://guix.gnu.org/manual/en/html_node/Common-Build-Options.html) succeeded. Have you put any thought into what a redo-build-system in Guix might look like? Since it advertises itself as a make replacement I do not expect that packages using it would be naturally compatible with gnu-build-system.

I do not see any contents in the commit message other than the header line and Change-ID. This might be related to the problem I explain in the next paragraph but please make sure that the commit message follows the [Change Log format](https://www.gnu.org/prep/standards/standards.html#Change-Logs).You can look at previous commits in the log to see relevant examples, patches that add new packages typically have simple messages. As a final note, I want to address some problems I ran into with the patch format. I'm not sure if the source of the problem is on your machine or if it has something to do with the server but I have previously downloaded patches from this server and they applied cleanly so I think it might have to do with your email client. The problems I ran into include the following:

1. Unchanged lines were prefixed with 2 spaces instead of 1. This caused `git apply` to report an error that the unchanged text could not be found.

2. There were 3 erroneous line breaks. This caused `git apply` to report that the patch was corrupt.

You can see the version of the patch I received with `wget https://issues.guix.gnu.org/issue/69661/attachment/0`.I manually modified the patch file so that it applies cleanly in order to perform the above review. Would it be possible for you to try sending the next revision using `git send-email` as [described by the manual](https://guix.gnu.org/manual/en/html_node/Sending-a-Patch-Series.html)? Note that while the section is titled "Sending a Patch Series" it also applies to sending single patches.

Change-Id: Ied142a7dc3e9baf9babdeff046f350e647a7a5cc

---

gnu/packages/build-tools.scm | 128 +++++++++++++++++++++++++++++++++++

1 file changed, 128 insertions(+)

Hi Skyler,

many thanks for your detailed review!

1) I sent right now a PATCH for: testing git send-email; fixing all

things in the review.

2) I don't consider this PATCH definitive, because: I can improve the

way I generate documentation; I'm not using the package enough in

production for being sure it is completely correct; a part of the

package is probably not optimal. I will send a candidate PATCH after

more testing and eventually your next review.

3) I will comment below to the questions of the past review...

DONE

Now I obtain

```

$ make && ./pre-inst-env guix lint redo-apenwarr

make all-recursive

]8;;file://think/home/mzan/tmp/guix/guix-cloned-channel/gnu/packages/build-tools.scm\gnu/packages/build-tools.scm:471:7]8;;\:

warning: failed to fetch Git repository for redo-apenwarr



]8;;file://think/home/mzan/tmp/guix/guix-cloned-channel/gnu/packages/build-tools.scm\gnu/packages/build-tools.scm:471:7]8;;\:

redo-apenwarr@0.42d: updater 'generic-git' failed to find upstream

releasesmake && ./pre-inst-env guix build -K redo-apenwarr

```

I have no idea how to fix them.

FIXED. Parallel builds and tests works correctly.

DONE

FIXED. You were right: it is not necessary.

I improved the package code and the comments. Probably now it is more

clear.

Put in other words: the redo-apenwarr installation script executes git

commands for querying the git repo, and for deriving the date of the

last commit. It uses this info for adding version/commit/date to the

installed application.

Apparently, after the Guix git-fetch phase, there is no anymore this

info, because there is no .git directory. So I generate this info

"manually".

This "patch" is not elegant, and I'm open to suggestions about the

correct way to handle this.

I double-checked all files. They are not header shebangs, but they are

shebangs in the middle of the code, because they will be added to some

generated files. So, I didn't changed this. It is correct.

DONE

FIXED

DONE

DONE

--- The output of the

I added `coreutils` also to the inputs of the package.

TODO: I need to test better in production, for seeing if it is working

all correctly, because despite it pass the tests, this tool interact

with the rest of the profile/system during execution.

I improved the description of the package. Previously I cited

"artifacts", but this term is not 100% correct, because the tool can be

used for producing many type of target files. It is more a

build/automation tool for rebuilding the chain of dependencies. For

example I'm using it for updating websites and so on.

So, it is not in competition with Guix, or other deterministic build

systems. It is a mix between a normal scripts, but with annotations for

having incremental updates. A corresponding Makefile would be a lot more

verbose and hard to comprehend, because it will use a bottom-up paradigm

in specifying dependencies.

Sadly, the main web-site of the tool is not describing it in a

marketing-free language, so I had to create a description from scratch.

Yes probably.

DONE. I hope this time it will be better. But, it is the first time I'm

using it.

Regards,

Massimo

On 3/12/24 08:39, Massimo Zaniboni wrote:

Ok, I'll add the "moreinfo" tag to this issue until you indicate that it is ready for merging.

I'm not seeing this on my machine, just a couple of warnings about trailing whitespaces. Since it failed to fetch a resource over the network I'm wondering if this was a temporary error that was preventing your machine from reaching the server? Let me know if you still get this error after successfully cloning the repository from the same machine (manually, with `git clone`).

This is very strange. I have used git commands in build phases before and there was no issue. I was using the same git-fetch origin type. The git folder *should* be there but when I remove the relevant snippet from the build phase and run with --keep-failed it is not. I'll look at this more closely when I have some time.

Looks like it worked! I was able to apply the patch locally without modification and I see your complete commit message now.

There are also a couple of small things about the new version of the patch that should be changed. In the PREFIX value for makeflags, "/out" should not be appended to #$output. This causes binaries to be installed to "out/bin" instead of "bin", so they are not actually in the path when users add the package to their profile. Also, since the build phase no longer uses the arguments given it can be a plain lambda instead of lambda*. You can include these changes when you send the next version that improves the documentation generation, no need to send an intermediate patch for these small changes.

user guix

tags 69661 + moreinfo

quit

On 3/12/24 18:52, Skyler Ferris wrote:

So, it turns out the reason that it worked for me previously is that I only used `git apply`, which is happy to apply patches even if they are not in a git repository, because the patch file itself has all the information it needs. So I guess this actually is expected.

I don't think that the manual substitution is necessarily a problem. But if you want to avoid it, perhaps using the [tagged release](https://github.com/apenwarr/redo/releases/tag/redo-0.42d)would work better? There is a comment in redo/version/gitvars.do that says that tarballs should have the correct data baked in.