Github-backup in 140 repository org backups nothing

[k8s-gpubuilder-markus]

github-backup prai-org -t _TOKEN_redacted --organization --private --all

2022-05-20T14:57:20.874: Backing up user prai-org to /home/mars/github_backup
2022-05-20T14:57:20.874: Requesting https://api.github.com/user?per_page=100&page=1
2022-05-20T14:57:21.094: Retrieving repositories
2022-05-20T14:57:21.095: Requesting https://api.github.com/orgs/prai-org/repos?per_page=100&page=1
2022-05-20T14:57:22.566: Requesting https://api.github.com/orgs/prai-org/repos?per_page=100&page=2
2022-05-20T14:57:23.295: Filtering repositories
2022-05-20T14:57:23.296: Backing up repositories
2022-05-20T14:57:23.296: Retrieving prai-org starred repositories
2022-05-20T14:57:23.296: Requesting https://api.github.com/users/prai-org/starred?per_page=100&page=1
2022-05-20T14:57:23.556: Writing 0 starred repositories to disk
2022-05-20T14:57:23.557: Retrieving prai-org watched repositories
2022-05-20T14:57:23.557: Requesting https://api.github.com/users/prai-org/subscriptions?per_page=100&page=1
2022-05-20T14:57:23.854: Writing 0 watched repositories to disk
2022-05-20T14:57:23.855: Retrieving prai-org-ai followers
2022-05-20T14:57:23.856: Requesting https://api.github.com/users/prai-org/followers?per_page=100&page=1
2022-05-20T14:57:24.117: Writing 0 followers to disk
2022-05-20T14:57:24.117: Retrieving prai-org following
2022-05-20T14:57:24.118: Requesting https://api.github.com/users/prai-org/following?per_page=100&page=1
2022-05-20T14:57:24.375: Writing 0 following to disk

This does a lot of nothing, a verbose flag would be important, why it is not backing up any of the 140 repositories.
A verbose flag (in my experience EVERY software needs it) would show what is being filtered and why it is just filtering everything, even when specifying --all

I'm a little confused, it says backing up user, but it actually means organization right? Because if I use the user instead of the org name, it can't find that.

[whwright]

It looks like there might be a bug with case sensitivity. Is the name of the organization all lowercase prai-org?

[fielding]

Just wanted to jump in and support @whwright's hypothesis. I was backing up my organizations and noticed the same thing, it is indeed case sensitive and matching how github shows it on the URL to the organization fixed the issue for me.

[matkoniecz]

I hit the same, turns out that I inverted streetcomplete and StreetComplete

Maybe it could show an error message "no such organization" or "no such repository" instead of showing logs that look like processing empty org?

[hkmaly]

Just for reference, MY problem of not backing up anything from organization was caused by using wrong token (personal instead of organization one). If someone had the same problem.

[Iamrodos]

This is a bug in the repository filtering logic. GitHub's API accepts usernames in any case (e.g., prai-org, PRAI-Org, PrAi-OrG) but always returns the canonical case in the response. When you type a username with different casing than GitHub's canonical form, the a case-sensitive string comparison causes all repositories to be filtered out.

How to reproduce:

Calling /user/repos returns all repos the authenticated user has access to. GitHub accepts any case in the call but returns the canonical case in the response:

curl -s -H "Authorization: token $TOKEN" https://api.github.com/user/repos | jq '.[0]'

# Returns repos with canonical case in owner.login:
{
  "name": "AdventOfCode",
  "owner": {
    "login": "Iamrodos"  // Capital I - canonical case from GitHub
  }
}

As the filer compares "Iamrodos" == "iamrodos" the result is false.

The bug:

At github_backup/github_backup.py:1590, the code does exact string matching:

if r.get("owner", {}).get("login") == args.user or r.get("is_starred"):

Since GitHub's API is not case-sensitive for logins, the tool's filter shouldn't be either.

History:

This bug was introduced in commit d362adb (January 2016) when the tool switched from /users/{user}/repos to /user/repos for private repository access. The /user/repos endpoint returns all repos the authenticated user can see (including organization repos), requiring filtering by owner. The filter comparison was inadvertently made case-sensitive.

Fix:

Change the comparison to be case-insensitive:

owner_login = r.get("owner", {}).get("login", "")
if owner_login.lower() == args.user.lower() or r.get("is_starred"):

Will create a PR with tests. I suspect many people have hit this issue, especially those with PascalCase usernames like me, aka "Iamrodos".