manifest: accept human-readable image sizes

[tomatotomata]

Fixes #2570

Summary

• let manifest --image-size use the existing datasizes.Size parser
• accept both raw byte counts and values with data-size units
• cover text marshaling and command flag parsing for both input forms

Testing

• go test ./pkg/datasizes
• GOOS=linux CGO_ENABLED=0 go test ./cmd/image-builder (cannot complete on this Windows host because github.com/proglottis/gpgme has no non-cgo build; the command package requires the project's Linux build dependencies)

[lucasgarfield]

Huh, a robot?

[supakeen]

Code works, thank you. One request for change in the review comments just for consistency.

It might be a bit harder but I'd like a commit message as well instead of just a subject line. It can include the usage (since users need to quote values-with-spaces).

---

Not 100% necessary in my opinion but maybe a manifest smoke test in main_test.go might make sense to ensure things don't blow up when passing this into manifest generation?

[supakeen]

The other unmarshallers wrap this is a new error (see for example line 39), let's do the same here.

[tomatotomata]

Fixed this in 2023bcb. UnmarshalText now wraps Parse errors with error decoding size: ..., matching the JSON and TOML paths, and TestSizeUnmarshalTextUnhappy locks in the context. go test ./pkg/datasizes passes. I left the unrelated worktree changes untouched. Let me know what you think.

[supakeen]

Huh, a robot?

Robots are taking over.

[tomatotomata]

I took the consistency request as adding a fuller command description, not only the one-line subject. Commit 8dda9ec adds Long and an explicit quoted image-size 1 GiB example, with a small help-text regression test. The existing parser and flag tests stay unchanged; let me know what you think.

[supakeen]

I took the consistency request as adding a fuller command description, not only the one-line subject. Commit 8dda9ec adds Long and an explicit quoted image-size 1 GiB example, with a small help-text regression test. The existing parser and flag tests stay unchanged; let me know what you think.

You didn't address the error wrapping comment: #2571 (comment) which is probably more important :)

[supakeen]

Thank you, looks good now. I'd prefer if you squash the fix commit(s) into the first commit but I won't block on that.

[lzap]

I am tired counting figures, love it.

[supakeen]

Needs a rebase @tomatotomata.

[tomatotomata]

I rebased the branch onto the current main and pushed 5db48e118. The focused parser and command tests still pass locally, and git diff --check is clean. The full command-package test remains Linux-only from this Windows host because the local checkout cannot build gpgme and the Unix statfs path. The new CI run is in progress. Let me know what you think once it finishes.