Update gh-md-toc, setting the -e flag

[mmattel]

Setting set -e to stop on first error. This is good bash practice...

BTW, I like this script a lot !!

[ekalinin]

Hi @mmattel, thanks for the PR and for the kind words!

I tested set -e against current master. It makes things worse in a few places, and it still doesn't catch the errors it's meant to catch:

1. Stdin input can silently lose the TOC. gh_toc_md2html runs on the left side of a pipeline, and that subshell inherits set -e. For example, say there is an unreadable token.txt next to the script and GH_TOC_TOKEN is not set. Before, printf '# Hello\n## World\n' | ./gh-md-toc - fell back to an unauthenticated request and printed the TOC. With set -e it prints an empty line and exits 0.
2. The network error message is lost in POSIX mode. When bash runs in POSIX mode (for example sh gh-md-toc README.md on macOS), command substitutions inherit set -e. A failing curl then exits before the network error check added in fix(md2html): check curl exit status for network errors #172 runs. Instead of "Parsing local markdown file requires access to github API" and exit 1, you get no message and exit 7. With --skip-header it also leaves a README.md~~ temp file behind.
3. --indent and --depth without a value now exit 1 with no message, because shift 2 fails.
4. Common failures still exit 0. There is no pipefail, and curl -s runs without --fail. So a network error on stdin input, a rate limit, or a 401 from a bad token still gives an empty TOC with exit code 0.

In --insert mode, set -e can also stop the script halfway. If writing the new TOC fails after sed -i has already cleared the old one, the script exits without telling the user where the backup is.

So I'd rather not add a global set -e. I think a targeted fix works better:

• check the XXNetworkErrorXX / XXRateLimitXX markers in the stdin path, the same way it's already done for local files
• possibly use curl --fail
• check that --indent and --depth were given a value
• add bats tests for these error paths

If you'd like to rework the PR in that direction, I'm happy to review it.

[ekalinin]

Requesting changes, since the problems from my comment above are still there on current master. A global set -e breaks two cases that work today:

1. Stdin input with an unreadable token.txt. Without GH_TOC_TOKEN, printf '# Hello\n## World\n' | ./gh-md-toc - prints an empty line and exits 0. Without set -e it falls back to an unauthenticated request and prints the TOC.
2. No network error message in POSIX mode. With sh gh-md-toc README.md (bash in POSIX mode, e.g. on macOS), a failing curl stops the script inside the command substitution, before the check from #172 runs. You get no message and exit 7 instead of "Parsing local markdown file requires access to github API" and exit 1. With --skip-header it also leaves a README.md~~ file behind.

The errors set -e was meant to catch are handled separately now: #151 adds the API error checks for stdin input, and #179 tracks the other error-handling bugs. If you'd like to rework this PR into targeted fixes, I'm happy to review it.