chore: de-duplicate shared helpers between the two testnet load tests

[olathedev]

Closes #97

What changed

scripts/testnet-load.sh and scripts/testnet-load-v2.sh each carried their own copy of nine helpers. Those now live in scripts/lib/load-test-common.sh, which both scripts source:

• log / log_success / log_error
• get_time / elapsed_time / avg_time
• gen_key, balance, invoke_contract
• plus STELLAR="stellar", which both set identically

network_id(), json_field(), the phase logic, and every contract-specific parameter stay where they were — the v2 script's helpers have no v1 counterpart, so there's nothing to share.

Split into three commits: the lib and CI change first (so the new file is linted from the commit that adds it), then one script each.

On the shellcheck glob

Answering the question from the issue thread: I kept the file at scripts/lib/ and updated the workflow rather than flattening it into scripts/, because the glob wasn't the only problem. shellcheck scripts/*.sh refuses to follow a source line at all without -x, and reports SC1091, which is fatal under the default severity — that would have failed CI even with the lib sitting directly in scripts/. So the step is now:

shellcheck -x scripts/*.sh scripts/lib/*.sh

-x makes shellcheck follow the source and check each script against the helper definitions it actually gets at runtime, instead of analysing the two halves in isolation; the explicit scripts/lib/*.sh keeps the lib checked standalone too. The # shellcheck source=SCRIPTDIR/lib/... directive on each source line resolves against the script's own directory, so the lint result doesn't depend on the workflow's working directory. Verified green at all three commits with shellcheck 0.10.0.

On set -euo pipefail and scoping

The lib deliberately has no shebang and no set line of its own. Sourcing runs in the caller's shell, which already sets -euo pipefail; re-setting them in the lib would silently re-enable an option for any future caller that had turned one off. It carries # shellcheck shell=bash instead so shellcheck still knows the dialect.

Every helper keeps its local declarations, so nothing new leaks into the callers' scope. The one shared name that isn't a function is STELLAR, which both scripts already defined to the same value.

The one helper that wasn't identical

balance() differed by exactly one token: v1 hardcoded --source load_deployer, v2 --source v2load_deployer. The identity name now lives in DEPLOYER_IDENTITY, set once per script in its configuration block, and the shared helper reads that. Each script's other deployer call sites use the variable too, so the name has one definition rather than being spelled out in five or six places.

NETWORK and DEPLOYER_IDENTITY must be set before the source line; both scripts set them in the config block immediately above it, and there's a comment on the source line saying so.

Verification

No behaviour change intended, and the helpers were checked rather than eyeballed: sourcing the pre-change definitions and the new shared ones and diffing declare -f for all nine gives byte-identical post-parse output in both directions, with balance's source argument as the sole expected difference. Both scripts pass bash -n, and shellcheck -x scripts/*.sh scripts/lib/*.sh is clean at each commit.

The scripts themselves deploy to live testnet, so I haven't run them end to end.

🤖 Generated with Claude Code

[collinsezedike]

The refactor itself is solid, verified byte-identical helper extraction, careful set -euo pipefail scoping, and the -x fix for shellcheck's SC1091 on the new source line all check out. But ci.yml's shellcheck command changed without updating three other places that document the equivalent local command:

• Makefile:15 still runs shellcheck scripts/*.sh
• CONTRIBUTING.md:194 still lists shellcheck scripts/*.sh as the raw local command
• README.md:116 still shows shellcheck scripts/*.sh labeled as "same checks CI runs"

None of these were touched by this PR, so running make check or copying the README/CONTRIBUTING command locally now fails with SC1091 on the new source line, even though CI itself passes with the -x flag. Please update all three to shellcheck -x scripts/*.sh scripts/lib/*.sh to match.