Continue running the go program after exit() in PHP

[thekid]

Currently, if a PHP script calls exit(), not only does the script terminate, but also the GO binary running it - see #31. This pull request changes this in a backwards-compatible way* such that calls to Exec() and Eval() will return an error which can be checked for.

Example

val, err := context.Eval(script);

if err != nil {
    if exit, ok := err.(*php.ExitError); ok {
        fmt.Printf("PHP exited with exit status %d\n", exit.Status)
    } else {
        fmt.Printf("Could not execute script %s: %v", script, err)
    }
}

*: ...instead of changing the return types, or introducing any global state in context, or adding new methods. The API is modeled a bit after os/exec

[thekid]

Hrm, this didn't have the desired effect of fixing the tests; unsure what is causing them to fail on Travis CI, then. However, these test failures have been around for a while looking at other pull requests.

[deuill]

Yeah, there have been some flaky tests that need fixing. I'll get on it.

[deuill]

Apologies for the delay in responding to these PRs, and thanks for doing the work. Looking now.

[deuill]

At a general level, I'm wondering if we should be storing more complex references to errors at the engine_context struct level, and have have introspected and returned in calls to Exec or Eval. Not bailing out when an error occurs is definitely desirable, but we may want to get more granular errors if we can?

[deuill]

I can see php_execute_script does a bunch of other stuff but ends up calling zend_execute_scripts. Was this changed because php_execute_script will truncate EG(exit_status)?

[deuill]

I think it may be possible to skip changing the signature of the version-specific functions (_context_eval and _context_exec) and just use EG(exit_status) here, since nothing else should be using the same context concurrently.

[deuill]

Needs to be zend_first_try too (my bad)?