Compatibility with podman

[katsuya-horiuchi]

First of all, thanks for the plugin. Running LSP from devcontainer was the feature I wanted for a while.

I use podman instead of docker, and starting devcontainer with :DevcontainersUp fails.

My LSP config for python looks something like this:

require("devcontainers").setup({
    docker_cmd = "podman",  -- Found this in config.lua
    log = { level = "trace" }
})

vim.lsp.config("pylsp", { cmd = require("devcontainers").lsp_cmd({ "pylsp" })})

When I run :DevcontainersUp, I get Starting devcontainer in /path_to_project: FAILED: code=1 status=nil
Checking log with :DevcontainersLog doesn't give much of a clue:

[2026-02-21 16:59:53.195270|TRACE] docker.events: subscribed: cache.invalidate
[2026-02-21 16:59:53.195698|TRACE] manager: devcontainer_up: /path_to_project
[2026-02-21 16:59:53.410814|TRACE] Command took 214.635 ms: devcontainer up --workspace-folder /path_to_project

Upon switching to docker (without docker_cmd in the option), the devcontainer starts and the plugin works as expected.

I'm not a lua guy, but the following might help debug the issue.

• When I open a python file, I get .../nvim/lazy/devcontainers.nvim/lua/devcontainers/init.lua:96: attempt to index local 'config' (a nil value)
• devcontainer-cli has --docker-path option, so without this plugin I have to do $ devcontainer --docker-path podman up to start devcontainer. Not sure if this is set properly in the plugin, though.

[jedrzejboczar]

I'm not entirely sure, but if you say you need to use devcontainer --docker-path podman, then maybe it will be enough to use:

require("devcontainers").setup({
    docker_cmd = "podman",
    devcontainers_cli_cmd = { "devcontainer", "--docker-path", "podman" },
})

When I open a python file, I get .../nvim/lazy/devcontainers.nvim/lua/devcontainers/init.lua:96: attempt to index local 'config' (a nil value)

About this one I don't know wny it would happen. Seems like vim.lsp did not pass the config argument to devcontainers.nvim's function. Maybe it's just that starting of devcontainer failed. Otherwise you may try debugging this function a bit more.

[katsuya-horiuchi]

I don't know how I didn't find devcontainers_cli_cmd option. Thanks.

About the other issue, updating the neovim and the plugin did the trick. However I encountered another issue, which I suspect is an issue with upstream.

The devcontainer starts successfully, but setting the client fails.
I get Client pylsp quit with exit code 127 and signal 0. (I'm trying to set up LSP for python with pylsp)
The log says rpc: server2client:on_exit: code=127 signal=0.

$ devcontainer --docker-path podman exec pylsp fails with 127 exit code, although bypassing devcontainer-cli (i.e. podman exec -it container_id pylsp) works.
Could you confirm that the plugin does something like this when setting a client?

[jedrzejboczar]

Yes, devcontainers.nvim basically just starts the container with devcontainer up --workspace-folder <dir> and then runs the lsp server with devcontainer --workspace-folder <dir> exec <user-cmd> and then keeps acting as man-in-the-middle to translate paths in LSP messages. If exec doesn't work in shell then it most likely won't work with this plugin.

You may try running pylsp with increased logging verbosity. I didn't dig into it, but there is -v (--verbose) flag (and --log-config which I'm not sure how it works). You can try running it in shell and then maybe

vim.lsp.config("pylsp", { cmd = require("devcontainers").lsp_cmd({ "pylsp", "-v" })})

[katsuya-horiuchi]

Perhaps the order of the command needs improving?
$ devcontainer exec --docker-path podman pylsp seems to work. (Notice that --docker-path is specified after exec)

Because I'm setting devcontainers_cli_cmd = { "devcontainer", "--docker-path", "podman" } in the config, I suspect exec is not placed in the right place.

[jedrzejboczar]

Ah, I see. So I guess I would need to add setting like:

require("devcontainers").setup {
    devcontainers_cli_cmd = { 'devcontainer' },
    devcontainers_cli_subcmd_args = { '--docker-path', 'podman' },
}

or

require("devcontainers").setup {
    devcontainers_cli_cmd = { 'devcontainer' },
    devcontainers_cli_subcmd_args = { 
        exec = { '--docker-path', 'podman' },
        up = { '--docker-path', 'podman' },
        -- and so on...
    },
}

This would then be used in lua/devcontainers/cli.lua. I can try to address this when I have some more time, or you can make a PR if you want to.

[katsuya-horiuchi]

I implemented a dirty solution, but it turned out netman.nvim wasn't compatible with podman either.

I'll look into netman.nvim, but in the meantime leave the solution below for reference.
For this solution, the only config required is docker_cmd = "podman".
It should work with docker as well, even though --docker-path isn't necessary for docker I believe.

diff --git a/lua/devcontainers/cli.lua b/lua/devcontainers/cli.lua
index c38760d..77ece29 100644
--- a/lua/devcontainers/cli.lua
+++ b/lua/devcontainers/cli.lua
@@ -120,7 +120,14 @@ function M.cmd(workspace_dir, subcommand, ...)
             return cmd
         end
     end
-    return utils.flatten(config.devcontainers_cli_cmd, subcommand, '--workspace-folder', workspace_dir, ...)
+
+    return utils.flatten({
+      config.devcontainers_cli_cmd,
+      subcommand,
+      '--docker-path', config.docker_cmd,
+      '--workspace-folder', workspace_dir,
+      ...
+    })
 end

 ---@param out vim.SystemCompleted
@@ -201,7 +208,10 @@ end
 ---@param opts? vim.SystemOpts
 ---@return { stdout: string, stderr: string }
 function M.exec(workspace_dir, cmd, opts)
-    local result = utils.system(M.cmd(workspace_dir, 'exec', unpack(cmd)), opts)
+    local result = utils.system(
+        M.cmd(workspace_dir, 'exec', '--docker-path', config.docker_cmd, unpack(cmd)),
+        opts
+    )
     if result.code ~= 0 then
         log.exception('exec failed for %s: stderr="%s" stdout="%s" signal=%s', workspace_dir, result.stderr, result.stdout, result.signal)
     end