Return an error on reading empty proc pid stat file

[pgimalac]

There is a race when reading /proc/<pid>/stat where the file can be opened while the process still exists but then it's gone when reading the content.
Depending on the underlying procfs, the file can just appear as empty in this case (ie. read just returns EOF directly), and the code happily tries to parse the file with the expected format, eventually resulting in a panic.

Testing whether the content is empty before calling splitProcStat should prevent most similar occurrences.

Note that more subtle cases can still happen, eg. if the file is long enough we might only read some of it, but Go's os.ReadFile uses a big enough buffer by default that it should be very rare.
Ideally the code should still check that the content has the expected format and not blindly access it.

Example

Below is a stack trace from an occurence:

panic: runtime error: slice bounds out of range [1:0]
goroutine 23291 [running]:
github.com/shirou/gopsutil/v4/process.splitProcStat({0xc0030e8a00, 0x0, 0x200})
	/pkg/mod/github.com/shirou/gopsutil/v4@v4.24.12/process/process_linux.go:1198 +0x29e
github.com/shirou/gopsutil/v4/process.(*Process).fillFromTIDStatWithContext(0x777657949328?, {0x5b72770, 0x9557240}, 0x30e4640?)
	/pkg/mod/github.com/shirou/gopsutil/v4@v4.24.12/process/process_linux.go:1053 +0x1de

splitProcStat is given an empty bytes slice, so LastIndexByte returns -1, and we index the empty slice starting at index 1 (-1 + 2), resulting in a panic.

	nameEnd := bytes.LastIndexByte(content, ')')
	restFields := strings.Fields(string(content[nameEnd+2:])) // +2 skip ') '

[pgimalac]

I used file is empty in the error as it's technically the only thing we can say about it, but in practice I believe it can only mean that the process has been stopped, so we could use a more explicit error instead.

[shirou]

Great catch! I wasn't aware of this stat file behavior. Thank you for this effective fix.

However, as you said, could you change the error message to indicate the likely cause?

[pgimalac]

@shirou done ! Thanks for the quick review !
I squashed the commits and CI is passing so should be good to merge

[shirou]

Thank you so much!