I've made live changes in production systems without testing them. If the production system is down, it's probably not going to get more down. Some of those changes are made on the live system and then back-ported into the release process. (Sometimes we aren't even sure which supporting index would help enough to come back, or which query to "neuter" to get site functionality mostly back. In cases like that, you might need to do development in production.) I've approved someone else shipping binaries built on a developer desktop to get a production site back functioning more quickly and then commit the changes are re-release from the build/deployment pipeline. We've pushed changes to prod without QA review. There are times to follow the measured, careful, prudent approach to development (most times) and there are other times where a meter of $10K/minute suggests that a lower latency change process is more appropriate and higher EV for the company.
That said, I've seen far more instances of no-reasonable-excuse events where code that was checked into master couldn't possibly compile, people doing an "svn resolve; svn commit" without actually resolving anything and checking in the ====== ++++++ conflict markers and both conflicted sections, etc.
I've made live changes in production systems without testing them. If the production system is down, it's probably not going to get more down. Some of those changes are made on the live system and then back-ported into the release process. (Sometimes we aren't even sure which supporting index would help enough to come back, or which query to "neuter" to get site functionality mostly back. In cases like that, you might need to do development in production.) I've approved someone else shipping binaries built on a developer desktop to get a production site back functioning more quickly and then commit the changes are re-release from the build/deployment pipeline. We've pushed changes to prod without QA review. There are times to follow the measured, careful, prudent approach to development (most times) and there are other times where a meter of $10K/minute suggests that a lower latency change process is more appropriate and higher EV for the company.
That said, I've seen far more instances of no-reasonable-excuse events where code that was checked into master couldn't possibly compile, people doing an "svn resolve; svn commit" without actually resolving anything and checking in the ====== ++++++ conflict markers and both conflicted sections, etc.
"There, I fixed it!"