sdk run was fixed in #201: user argv was spliced into the container's shell script with a bare join(" "), so the container shell re-split and re-expanded it. The same splice is still raw at three other sites, found by @jetm reviewing that PR:
src/commands/sdk/dnf.rs:115src/commands/ext/dnf.rs:358src/commands/runtime/dnf.rs:303
All three are self.command.join(" ") on user-supplied argv, the identical shape.
The fix is now a one-liner each — #201 moved shell_escape to utils::shell and added join_argv — but it carries the same user-visible behavior change as #201 did: an argument containing a $VAR, a glob or a && stops being interpreted by the container shell and starts arriving literally. That is worth deciding deliberately for the dnf subcommands rather than folding into a sdk run fix, which is why it was left out of #201.
Wants a CHANGELOG entry and the same argv-boundary test per site.
sdk runwas fixed in #201: user argv was spliced into the container's shell script with a barejoin(" "), so the container shell re-split and re-expanded it. The same splice is still raw at three other sites, found by @jetm reviewing that PR:src/commands/sdk/dnf.rs:115src/commands/ext/dnf.rs:358src/commands/runtime/dnf.rs:303All three are
self.command.join(" ")on user-supplied argv, the identical shape.The fix is now a one-liner each — #201 moved
shell_escapetoutils::shelland addedjoin_argv— but it carries the same user-visible behavior change as #201 did: an argument containing a$VAR, a glob or a&&stops being interpreted by the container shell and starts arriving literally. That is worth deciding deliberately for thednfsubcommands rather than folding into asdk runfix, which is why it was left out of #201.Wants a CHANGELOG entry and the same argv-boundary test per site.