无法分叉java可执行文件,JVM：如何定义Java可执行文件的退出代码？

I'm looking for a definition of possible exit codes of the java executable:

(How) can I tell whether the exit code is of the executed Java process or of the VM itself?

Example: On Windows, java -badoption returns 1; java Main with Main being a valid class may return 1 as well.

Are there any VM options I could use to make exit codes more meaningful? E.g. to distinguish between both types of exit codes?

If I know that the exit code is not from my Java process (which only returns 0), what do non-zero exit codes mean?

On Windows, I'm frequently seeing -1 and 1. As these are reported through an automated bug-reporting facility, I can't see any error messages. I only have the exit code and need to interpret it.

Are the exit codes platform-dependent?

解决方案

Meaning of Exit Codes (in general)

ISO/IEC 9899 (Programming Languages: C), ISO/IEC 9945 (IEEE 1003, POSIX)

There is nothing which is specifically defined for Java / JVM. The meaning of these exit codes is defined by ISO/IEC 9899 (Programming Languages: C, for example 7.20.4.3 The exit function of ISO/IEC 9899:TC3), ISO/IEC 9945 (IEEE 1003, POSIX) and similar specifications, and it's always like this:

0 means success

Any other value means failure

This is used by shell environments (sh, bash, cmd.exe, make etc. etc.) to determine whether a program exited successfully (0) or there was an error (not 0).

It is strongly recommended to use only 0 for success. Programs which use values other than 0 for success cause headache to the authors of shell scripts, Makefiles and alike.

It's up to the individual program to define additional semantics of different failure values. However, I think -1 is not such a good idea. The best thing would be to use EXIT_SUCCESS and EXIT_FAILURE, however, these definitions from have no corresponding counterparts in Java. ISO/IEC 9899:2011 does not describe how EXIT_FAILURE and EXIT_SUCCESS should be defined. However, it defines 0 as having the same semantics as EXIT_SUCCESS, and I do not know of any system which does something else than the following, which therefore is the best thing a Java programmer can assume:

#define EXIT_SUCCESS 0

#define EXIT_FAILURE 1

So, I'd use 0 for success, 1 for failure, and values distinct from 1 if it is important to differentiate between different types of failure. A good example is the grep command. From its man page (GNU grep):

EXIT STATUS

The exit status is 0 if selected lines are found, and 1 if not found. If an error occurred the exit status is 2. (Note: POSIX error handling code should check for '2' or greater.)

I'd not use -1 because it sets all bits, and depending on the environment, -1 might actually accidentally generate additonal bogus information, like claiming that signals were raised and alike. Actually, to be portable, the exit code should be within the range [0..127], unless you really really really know what you're doing.

For example, on most POSIX systems, the exit code will be truncated to 8 bits, and the semantics of an exit code above 127 is that the exit was caused by a signal.

BSD

BSD has tried to supply a few more exit codes, which are available in /usr/include/sysexits.h (man page). They are like this:

64: command line usage error

65: data format error

66: cannot open input

67: addressee unknown

68: host name unknown

69: service unavailable

70: internal software error

71: system error (i.e. can't fork)

72: critical OS file missing

73: can't create (user) output file

74: input/output error

75: temp failure; user is invited to retry

76: remote error in protocol

77: permission denied

78: configuration error

In the absence of any other meaningful standard, I think the best thing we can do is use them.

Interpretation of JVM-generated exit values

JVM on POSIX

On UNIX, you can obtain the signal by masking out the lower 7 bits (subtract 128) from the exit value, which in most shells can be queried using $?.

SIGTERM -> VM: 143 SIGTERM

SIGSEGV -> VM: 134 SIGABRT (because the VM handles SIGSEGV to write the hs_err file and then calls abort()).

Comparison to "normal" programs:

SIGSEGV -> prog: 139 SIGSEGV

This behavior is not explicitly specified for the JVM, it is just the normal expected behavior of programs in a POSIX environment. Even the SIGABRT instead of SIGSEGV is more or less expected, given that the JVM wants to handle SIGSEGV on its own, writing a more specific crash dump than a normal core file.

JVM on Windows

TODO