FIT2109 Chap.11 Metaprogramming and Automation
Metaprogramming and Automation
Metaprogramming and Automation begins with a stable task contract. State the inputs, ordered transformations, expected artifacts, failure conditions and points that still require judgement. Automation should remove repetition while preserving the evidence needed to review decisions.
Generated output needs clear ownership and provenance.
Build a complete candidate, validate it and replace an owned artifact atomically instead of appending blindly. Test the second run as well as the first. Unchanged inputs should preserve the same logical result rather than duplicate rows, configuration blocks or generated sections. A required-stage failure must leave the prior valid artifact untouched and return a visible failure.
What this chapter covers
- 01
Manual task contracts before implementation
- 02
Explicit input and output ownership
- 03
Generated artifacts separated from hand-edited content
- 04
Idempotent behaviour across repeated runs
- 05
Atomic construction and promotion
- 06
Failure status that cannot resemble success
- 07
Logs that support a concrete next action
- 08
Human checkpoints for ambiguous or policy-sensitive decisions
Worked example · free
Make a generated reference idempotent
- +1Define the command definitions as the source of truth and the reference as a generated artifact.
- +1Build the complete next reference in memory or a temporary destination.
- +1Validate that every expected command appears exactly once.
- +1Mark the output as generated and point readers to the source of truth.
- +1Replace the destination atomically only after validation succeeds.
- +1Return non-zero and keep the previous valid artifact if a required scan fails.
- +1Run the generator twice with unchanged inputs and compare the results.
- +1Keep hand-written guidance in a separate owned file or documented protected region.
Key terms
- Task contract
- A statement of inputs, operations, outputs, failure states and review points that defines what an automation must accomplish.
- Generated artifact
- Output owned by a repeatable process rather than by ordinary hand editing. Its source and regeneration route should be visible.
- Idempotence
- The property that rerunning an operation with unchanged inputs preserves the same logical result without accumulating duplicates or drift.
- Atomic replacement
- Constructing a complete candidate separately and replacing the destination only after validation, so partial work never appears complete.
- Promotion gate
- A required verification boundary that a candidate artifact must pass before receiving a stable name or publication status.
- Human checkpoint
- A visible decision point reserved for ambiguous intent, safety, policy or contextual judgement that automation should not silently resolve.
Metaprogramming and Automation FAQ
What should be automated first?
Choose a repeated task whose inputs, outputs and success checks can be stated clearly. Stabilise the manual route before encoding it so the tool does not preserve an unclear or unsafe process.
Why test an automation twice?
The first run often starts from an empty destination. The second reveals duplicate appends, stale temporary state and unclear ownership. It is direct evidence for idempotent behaviour.
How should a failed generation behave?
It should return a visible failure, preserve the previous valid artifact and retain enough concise diagnostics to identify the failing source or stage. Partial output should not be promoted.
Which decisions should remain human-reviewed?
Keep ambiguous intent, safety boundaries, policy choices and publication decisions visible. Automation can collect evidence, repeat checks and format results without silently resolving every contextual judgement.
Assessment move
Choose one repeatable task such as generating a command reference or running a verification matrix. Write its contract before writing code. Identify the source of truth, candidate artifact, validation checks and stable destination. Implement the route so required failures return non-zero and leave the previous valid output unchanged. Run it twice with unchanged inputs and compare results.
Introduce one failure midway and confirm that no partial artifact is promoted. Review logs for usefulness and remove secrets or irrelevant noise. Next list decisions that the tool can make deterministically and decisions that still need human review. Explain why each boundary exists. This practice connects shell status, Git ownership, container verification and data reconciliation into one safe automation design.
Add a content fingerprint and a human-readable diff to the promotion step so an operator can see whether the candidate changed. Run two copies against separate temporary destinations to expose hidden global state. Finally, hand the instructions to another person or a fresh shell and observe which undocumented assumption prevents completion; convert that assumption into an explicit precondition or automated check.
Working through Metaprogramming and Automation in FIT2109? Sia is AskSia’s AI Computer Science tutor — ask any FIT2109 Metaprogramming and Automation question and get a clear, step-by-step explanation grounded in how FIT2109 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
