feat(pro report): add --bundle-id and --path to software-installs

[ktn-jamf]

Closes #325.

pro report software-installs reported three columns: title, version, device_count. This adds --bundle-id and --path, which report an application's bundle identifier and its install path.

Both fields were already on the wire. The ComputerApplication schema declares bundleId and path, and the function already read path to feed isSystemApp. No spec change and no regeneration are needed.

Each flag extends the grouping key

This is the correctness point of the change. One title and version can map to several bundle identifiers or several install paths — a repackaged application, or the same application installed in two locations. A display-only column would print one arbitrary member of the group and hide the rest, with nothing in the output to reveal it. So each flag joins the aggregation key:

$ jamf-cli pro report software-installs --title Zoom --bundle-id -o table
 BUNDLE_ID                DEVICE_COUNT  TITLE  VERSION
 com.example.zoom-repack  1             Zoom   5.0
 us.zoom.xos              2             Zoom   5.0

Without the flag those three installs are one row reading 3 for Zoom 5.0.

Column order is alphabetical. The table formatter reads its column set from sortedKeys(rows[0]), which floats only id and name, so every table in this CLI orders the rest that way. Use --select title,version,bundle_id,device_count for a different order.

softwareKey gains bundleID and path, left at the zero value when the flag is off, so the counting loop keys itself correctly in all four flag combinations with no branching. Sorting gained tie-breakers on bundle identifier then path, because a title and version can now span rows.

The default output is unchanged

The row builder gates each optional key on the flag, never on whether the value is empty. A table's columns are the keys of its first row, so an emptiness gate would make the column set depend on sort order, and an application with no bundleId on the wire would drop the column for every row. Both flags are off by default, so anything parsing the current three columns is unaffected. A test asserts that the default rows carry exactly those three keys and no more.

Flag names

Kebab-case, matching --include-system and every other flag in this CLI. The issue text asks for --bundle_id; snake_case would be the only such flag in the binary. --path covers the --file_path request in the issue comment.

Tests

Four cases beside the existing four:

• TestRunReportSoftwareInstalls_BundleIDExtendsTheGroupingKey
• TestRunReportSoftwareInstalls_PathExtendsTheGroupingKey
• TestRunReportSoftwareInstalls_DefaultRowsCarryOnlyTheThreeOriginalKeys
• TestRunReportSoftwareInstalls_BundleIDAndPathTogether

The first two are the ones that fail if the grouping is ever reduced to a display column. That was confirmed by mutation: discarding the two key assignments makes exactly the grouping tests fail, and no others.

make test passes in full with zero failures. go vet ./internal/commands/ is clean. make verify-site and make verify-site-output both pass, since the change touches docs/site.

Second commit, from review

A deep review found one confirmed defect and two real test gaps.

Both flags are boolean and both names invite a value, so --path /Applications/Foo.app set the flag and left the path as a positional argument, which the command discarded in silence and then reported the whole fleet. The operator got fleet-wide output for what they had typed as a filter. Args: cobra.NoArgs refuses it: cobra reports it as unknown command, which ClassifyError already maps to exit 2, and ValidateArgs runs ahead of PersistentPreRunE, so the refusal lands before any credential resolves or any request is sent.

The hole itself is repo-wide and pre-existing — pro report inventory-summary junkarg and pro categories list junkarg swallow a positional the same way — so the fix is scoped to the one command whose new flag names make a latent hole reachable. The general case deserves its own change.

The two test gaps: nothing in this file asserted row order at all, so the new tie-breakers were unobservable, and neither empty-value case was covered. TestRunReportSoftwareInstalls_RowOrderFollowsEveryComparatorLevel puts a tie at each of the four comparator levels and pins the full sequence; flipping any level in turn fails it. Two further tests cover an application the wire reports with no bundleId and one with no path, both of which still get a row carrying the column, because the row builder gates on the flag rather than on the value. Mutating either gate to test the value fails them.

The two flags also joined docs/site/examples.json. No test validates that file, so it was checked by parsing it, comparing its key set against every other entry, and running both documented commands.

The command needs a live Jamf Pro tenant, so the fixture-client tests are the runtime evidence for the aggregation. The rendered --help was checked on the built binary.

make lint was not run. The golangci-lint on the development machine is built with go1.26.2 while go.mod targets 1.27.0, so it aborts at config load before it reads any file. CI installs its own version and is unaffected.

🤖 Generated with Claude Code