Server not receiving information to compute hash

[b-rodrigues]

The server is currently not receiving the right info required to compute the hash if nix is not present on the system. The right result is returned via curl though:

 curl -X GET "https://git2nixsha.dev/hash?repo_url=https%3A%2F%2Fgithub.com%2Feliocamp%2FmetR&commit=1dd5d391d5da6a80fde03301671aea5582643914" -H  "accept: */*"

but unit tests on a system without Nix fail:

── Error ('test-fetchers.R:131:3'): Test fetchgits works when PR is provided in a remote package, but does not use it ──
Error: Error: Request `curl::curl_fetch_memory(url = 'https://git2nixsha.dev/hash?repo_url=&commit=')` failed:
 HTTP response code said error [git2nixsha.dev]: The requested URL returned error: 500
If it's a GitHub repo, check the url and commit.
Are these correct? If it's an archived CRAN package, check the name
of the package and the version number.

see: https://github.com/ropensci/rix/actions/runs/13562423321/job/37911046377

fyi @mihem perhaps you see where this is coming from?

[b-rodrigues]

For example, this fails:

 fetchgit(
      list(
        package_name = "CSFAtlasTools",
        repo_url = "https://github.com/mihem/CSFAtlasTools",
        commit = "ac4a34d39812e8b8fa5746b63df4cf321a13b7a7"
      )
    )

but not the one where housing is fetched, so it seems that packages with remote deps are not being correctly handled

[b-rodrigues]

Using the debugger:

• with the local nix install, repo_url is first CSFAtlasTools, and then get_sri_hash_deps gets called again and repo_url is datathin
• using the api server, when get_sri_hash_deps gets called again, this time repo_url is NULL

so we need to understand why that is

[mihem]

In general, I think the issue is that we never checked nix_hash_online. I actually never changed code there but maybe indirectly broke something there or it never worked.

Yeah I found debugging really helpful, at work right now, so cannot do this (maybe this evening)

But yes that must be it, why is repo_url NULL.

[b-rodrigues]

nix_hash_online did work, but I think I remember I turned off the fetchers tests with the second rix release on CRAN. I think a test was failing on CRAN due to some changes in a repo if I remember correctly, so in the meantime, the changes to the code that introduced between that release and now broke something. It seems as if the arguments are not being explicitly passed, so maybe it’s quite an easy fix

[b-rodrigues]

ok so the plot thickens; trying this directly with curl:

curl -X GET "https://git2nixsha.dev/hash?repo_url=https://github.com/mihem/CSFAtlasTools&commit=ac4a34d39812e8b8fa5746b63df4cf321a13b7a7" -H  "accept: */*"

results in the following:

{
    "sri_hash": ["sha256-w2yEeAMI/+zLgYeEfPRNzf25q35l0KSY8kiJXgeMBwI="],
    "deps": {
        "package": ["CSFAtlasTools"],
        "imports": ["dplyr", "glue", "readr", "ggplot2", "tidyr", "RColorBrewer", "bestNormalize", "pheatmap", "recipes", "tibble", "viridis", "broom", "WRS2", "Seurat", "abind", "mclust", "tune", "yardstick", "ggsignif"],
        "remotes": [{
            "package_name": ["datathin"],
            "repo_url": ["https://github.com/anna-neufeld/datathin"],
            "commit": [
                ["58eb154609365fa7301ea0fa397fbf04dd8c28ed"]
            ]
        }]
    }
}

[b-rodrigues]

also, to keep in mind, nix_hash_online calls nix_hash on the server: https://github.com/b-rodrigues/git2nixsha/blob/main/git2nixsha.R

[b-rodrigues]

ok so the plot thickens; trying this directly with curl:

curl -X GET "https://git2nixsha.dev/hash?repo_url=https://github.com/mihem/CSFAtlasTools&commit=ac4a34d39812e8b8fa5746b63df4cf321a13b7a7" -H  "accept: */*"

results in the following:

{
    "sri_hash": ["sha256-w2yEeAMI/+zLgYeEfPRNzf25q35l0KSY8kiJXgeMBwI="],
    "deps": {
        "package": ["CSFAtlasTools"],
        "imports": ["dplyr", "glue", "readr", "ggplot2", "tidyr", "RColorBrewer", "bestNormalize", "pheatmap", "recipes", "tibble", "viridis", "broom", "WRS2", "Seurat", "abind", "mclust", "tune", "yardstick", "ggsignif"],
        "remotes": [{
            "package_name": ["datathin"],
            "repo_url": ["https://github.com/anna-neufeld/datathin"],
            "commit": [
                ["58eb154609365fa7301ea0fa397fbf04dd8c28ed"]
            ]
        }]
    }
}

which I believe is right, but probably at some point we try to grab repo_url using x$repo_url where in this case it should be x$remotes$repo_url probably something like this. We might need to restructure the vector, use x$repo_url if repo_url exists, and x$remotes$repo_url if not.

[b-rodrigues]

So, the remotes object when using Nix locally:

it’s a list.

and this is the remotes object when using the server:

it’s a data frame. So probably the way the object is returned needs to be fixed.

[mihem]

yeah that's exactly what i meant here
#424 (comment)

[b-rodrigues]

yeah that's exactly what i meant here #424 (comment)

I opened a #431 dealing with this, it is better now, it falls back to head, but at least the response from the server seems to be correct

[mihem]

Nice, btw, not sure there is something similar in ESS, I can really recommend the debugging in VSCode R https://github.com/ManuelHentschel/vscDebugger
It's super helpful when you can see the call chain, the variables of the environments and so on (which i miss when just using debugonce althoguh this is still super helpful), or how do you debug things?

[b-rodrigues]

i simply use browser(), but you’re right I should explore more advanced tools

[mihem]

yes agree that's also very helpful, but in this case with all the different environemnts, scopes, variables, this debugging tool was really helpful.
Yeah, you should finally test vscode R (and maybe ESS has something similar), see how nice this looks: