[metacling] Prevent nullptr access if compiler instance was not created

[ferdymercury]

This Pull request:

Changes or fixes:

exit with error1 code instead of crashing

Fixes https://its.cern.ch/jira/browse/ROOT-10239

Checklist:

• tested changes locally
• updated the docs (if necessary)

[github-actions]

Test Results

    18 files      18 suites   4d 0h 5m 9s ⏱️
 2 678 tests  2 677 ✅ 0 💤 1 ❌
46 342 runs  46 337 ✅ 4 💤 1 ❌

For more details on these failures, see this check.

Results for commit 5772649.

♻️ This comment has been updated with latest results.

[dpiparo]

this is good improvement: thanks! I was wondering whether exiting is the best course of actions or if an exception or a clear error message could be of help. Now that the location in the code has been identified, I can also re-open a PR including the commit included in this one and enhance the visibility of the problem for the user (which is a simple thing to do, at this point...)

[ferdymercury]

I tried first throwing a runtime_exception, but the problem is that this class has an exception handler which auto-dumps a full stack trace onto stdout, which is exactly what we didn't want :S So exit(1) seemed cleaner since the error message was already printed before. A middle ground would be to add another "cout" before the exit(1);

[dpiparo]

Ah, I see. Thanks. Can we maybe go for a last "cerr" before the exit? Then for me the PR is ready to merge for me.