Java Forum / General / May 2006
Runtime.exec(String[]) Doesn't Always Work, bBut Runtime.exec(String) Does
Hal Vaughan - 21 May 2006 20:49 GMT I've included the method I'm having trouble with at the end of the post.
I was using Runtime.getRuntime().exec(String) to execute programs and had no problem with it at all, but since I had to include an argument, I was worried about problems with spaces in program names, so I switched to .exec(String[]) instead of .exec(String). I figure this way when it turns out I have to run a program with a space in the filename, there should be no problem.
When I was specifying the program with only a String, every program I would attempt would run. Now that I'm using String[], I have times where I run one program and it works, then run the same program, with the same arguments, a few seconds later, and waitFor() returns immediately and I get no data from the ErrorStream or InputStream.
I tried, as an experiment, to put quotation marks around the filename in a string so it would look like this:
"/usr/bin/myprogram" -myargument
And that would not work -- the programs would not run at all.
Also as an experiment, in the spot I marked with a comment, I tried a sleep() call, just in case waiting had any effect. When I had it wait 100 milliseconds, every time I ran a program, I got data from the appropriate InputStream or ErrorStream, but not all the data I should have (it was lopped off at the beginning). I tried shorter times for sleep() and with 10 milliseconds, no programs ran.
So why is it that sometimes the program runs and I get input and sometimes it doesn't? My best guess is that the program always runs, but that there might be a timing issue that keeps me from reading the streams quickly enough, but that doesn't sound likely.
Any help is appreciated.
Thanks!
Hal
------Code method----- public boolean runFile() { stopFlag = false; System.out.println("DEBUG: Starting to run file, Command: " + runLine); try { Process procRun = Runtime.getRuntime().exec(cmdInfo); iRead = new BufferedReader (new InputStreamReader(procRun.getInputStream())); eRead = new BufferedReader (new InputStreamReader(procRun.getErrorStream())); System.out.println("\tDEBUG: Buffers made"); //Tried to use Thread.sleep() here with different timings //Thread to read Standard output //Only exception thrown by BufferedReader.readLine() is EOF, which //means we're at the end, so there's no need to do anything with it. new Thread( new Runnable() { public void run() { String sRead; try { while ((sRead = iRead.readLine()) != null) { System.out.println("DEBUG: Standard:" + sRead); outData = outData + sRead; if (stopFlag) break; } } catch (Exception eofE) {} } } ).start(); //Thread to read Error output new Thread( new Runnable() { public void run() { String sRead; try { while ((sRead = eRead.readLine()) != null) { System.out.println("DEBUG: Error:" + sRead); outError = outError + sRead; if (stopFlag) break; } } catch (Exception eofE) {} } } ).start(); procRun.waitFor(); stopFlag = true; System.out.println("\tDEBUG: Run is complete"); iStream.close(); eStream.close(); } catch (Exception e) { System.err.println("Could not execute IDFile: " + runLine); return false; } System.out.println("\tDEBUG: Done running file"); return true; }
Thomas Hawtin - 21 May 2006 21:02 GMT > Also as an experiment, in the spot I marked with a comment, I tried a > sleep() call, just in case waiting had any effect. When I had it wait 100 > milliseconds, every time I ran a program, I got data from the appropriate > InputStream or ErrorStream, but not all the data I should have (it was > lopped off at the beginning). I tried shorter times for sleep() and with > 10 milliseconds, no programs ran. The start of the data missing? That's very odd. It should drop anything from the streams.
> procRun.waitFor(); > stopFlag = true; You are instructing you child threads to stop before they have necessarily read all of the data. (I assume stopFlag is volatile.) They wont read the flag if there is no data to read and the streams are not closed. A better idea would be to jsut let them complete normally.
> System.out.println("\tDEBUG: Run is complete"); > iStream.close(); > eStream.close(); Now you are attempting to close the streams, before they are necessarily read.
Tom Hawtin
 Signature Unemployed English Java programmer http://jroller.com/page/tackline/
Hal Vaughan - 21 May 2006 21:15 GMT >> Also as an experiment, in the spot I marked with a comment, I tried a >> sleep() call, just in case waiting had any effect. When I had it wait [quoted text clipped - 6 lines] > The start of the data missing? That's very odd. It should drop anything > from the streams. I tried again, with a sleep() call. I may have gotten some output files mixed up, but it DID lose data from the streams. It may not have been from the beginning, but I'm pretty sure it was.
>> procRun.waitFor(); >> stopFlag = true; [quoted text clipped - 3 lines] > wont read the flag if there is no data to read and the streams are not > closed. A better idea would be to jsut let them complete normally. So would I be better off just reading them continually until they throw an EOFException? It seems that would be the best if that is the case.
> > System.out.println("\tDEBUG: Run is complete"); > > iStream.close(); > > eStream.close(); > > Now you are attempting to close the streams, before they are necessarily > read. Is that so even if I'm closing them AFTER I'm using waitFor()?
Thanks!
Hal
Rhino - 21 May 2006 23:29 GMT >>> Also as an experiment, in the spot I marked with a comment, I tried a >>> sleep() call, just in case waiting had any effect. When I had it wait [quoted text clipped - 34 lines] > > Thanks! I found this article quite helpful when I was messing with Runtime.exec() a couple of years back. It may help you sort out your current problem or at least understand better how it works: http://www.javaworld.com/javaworld/jw-12-2000/jw-1229-traps_p.html.
-- Rhino
Hal Vaughan - 21 May 2006 23:26 GMT >> Also as an experiment, in the spot I marked with a comment, I tried a >> sleep() call, just in case waiting had any effect. When I had it wait [quoted text clipped - 23 lines] > > Tom Hawtin I made changes based on these suggestions and it works now,with only one issue that I fixed with a hack I'm not thrilled with.
The streams are not closed at the end of the routine, but within the threads that read from the streams. When reading from a stream throws an error (and the only documented error they throw is EOF), the stream is closed.
I still use waitFor() at the end to determine when the program finishes running, but after that, I wait 100 milliseconds, to allow any extra data to be pulled from the streams. Then I return. That's the hack I'm worried about. I can't wait too long, since I'm going through a number of files, but without waiting, I sometimes don't get any data from the streams. I'd think there should be a better way to make sure all the output from the program has been received, but when the streams are done (i.e. all the program output has completed), they don't always throw an EOF.
I tried setting flags when each stream threw an error and waiting until both streams had thrown an EOFException, but when I did that, it would freeze and at least one of the streams never closed. If there is another way to handle this, other than waiting for some time to allow the streams to receive all the data, I'd really like to hear about it.
Thanks for any help and suggestions.
Hal
Gordon Beaton - 22 May 2006 08:31 GMT > I still use waitFor() at the end to determine when the program > finishes running, but after that, I wait 100 milliseconds, to allow > any extra data to be pulled from the streams. Then I return. > That's the hack I'm worried about. Simply read until readLine() returns null. That's BufferedReader's way of indicating EOF.
/gordon
 Signature [ do not email me copies of your followups ] g o r d o n + n e w s @ b a l d e r 1 3 . s e
Hal Vaughan - 22 May 2006 15:32 GMT >> I still use waitFor() at the end to determine when the program >> finishes running, but after that, I wait 100 milliseconds, to allow [quoted text clipped - 5 lines] > > /gordon If the process is still running but hasn't printed anything to output for a while, what does readLine() return? An empty string instead of a null?
Thanks for the pointer. I'll add that to the routine.
An extra note: I did searches and found Sun's "official" method for reading from a process and it does not take a lot of what has come up in this thread into account.
Hal
Gordon Beaton - 22 May 2006 17:49 GMT > If the process is still running but hasn't printed anything to > output for a while, what does readLine() return? An empty string > instead of a null? It's much simpler than that: readLine() *doesn't* return until there *is* a line of text to return. At the end of the stream, it returns null.
/gordon
 Signature [ do not email me copies of your followups ] g o r d o n + n e w s @ b a l d e r 1 3 . s e
Thomas Hawtin - 22 May 2006 11:10 GMT > The streams are not closed at the end of the routine, but within the threads > that read from the streams. When reading from a stream throws an error > (and the only documented error they throw is EOF), the stream is closed. Rather than trying to sleep, I'd suggest using Thread.join. You actually only need one additional thread, as you are going to wait for the process anyway.
I don't see any documentation that states that other types of IOException are not thrown.
BTW: If your process returns many lines, the Java code will become increasingly slow. Repeated string concatenation is bad - use StringBuilder (or StringBuffer).
Tom Hawtin
 Signature Unemployed English Java programmer http://jroller.com/page/tackline/
Hal Vaughan - 22 May 2006 15:39 GMT >> The streams are not closed at the end of the routine, but within the >> threads [quoted text clipped - 4 lines] > only need one additional thread, as you are going to wait for the > process anyway. I had never thought of that -- I'm not familiar with it and the API docs don't go into detail, so I'll read up on it.
> I don't see any documentation that states that other types of > IOException are not thrown. Thanks. That confirms what I could find..
> BTW: If your process returns many lines, the Java code will become > increasingly slow. Repeated string concatenation is bad - use > StringBuilder (or StringBuffer). > > Tom Hawtin I'm not too worried. Basically, this is part of a file finder. It looks for files to identify them. Then it either reads the file in to find identifying text, or runs it to check the output. That means usually the program is run with something like "--help" or "-version" or something similar and all I'm doing is reading in the output to compare it against a version string.
Hal
Gordon Beaton - 22 May 2006 08:27 GMT > When I was specifying the program with only a String, every program > I would attempt would run. Now that I'm using String[], I have times [quoted text clipped - 8 lines] > > And that would not work -- the programs would not run at all. exec(String) simply tokenizes the command and invokes exec(String[]). So if exec(String) works but exec(String[]) doesn't, any problems with the latter are most likely due to your tokenization of the command line.
exec(String[]) is, as you correctly observe, the right choice when your command contains *significant* whitespace, because the StringTokenizer used by exec(String) is not particularly smart.
The rule is simple: each single component in the command line, i.e. the command name, an option flag, an argument, etc, is one String in the command array.
Don't use any quotes or escape characters, because there is no shell to further interpret the command line. Any such characters will be treated literally by the system, so unless your program really is spelled with those characters it won't be found, or the arguments will be meaningless, etc.
> So why is it that sometimes the program runs and I get input and > sometimes it doesn't? My best guess is that the program always runs, > but that there might be a timing issue that keeps me from reading > the streams quickly enough, but that doesn't sound likely. There should not be any timing issues. Data written to either of the streams stays there until you read it. And if you don't read it, the child will eventually hang waiting for you to do so, as you've already discovered. And you won't get EOF until you've emptied the streams and the child has exited (or closed the streams).
I don't see BufferedReader.ready() or InputStream.available() in your code, but use of those could give the results you describe (avoid them).
However if waitFor() returns, that is a string indication that your program has exited. What is returned by Process.exitValue()? What does your command line look like?
/gordon
 Signature [ do not email me copies of your followups ] g o r d o n + n e w s @ b a l d e r 1 3 . s e
Hal Vaughan - 22 May 2006 15:45 GMT >> When I was specifying the program with only a String, every program >> I would attempt would run. Now that I'm using String[], I have times [quoted text clipped - 27 lines] > spelled with those characters it won't be found, or the arguments will > be meaningless, etc. Thanks for clearing that up. I had found out about the quotes the hard way, but wasn't sure of the why behind it and didn't realize the String was tokenized, but that also confirms my thought that spaces in the file name would be trouble. Thanks for the details and explicit background on this.
>> So why is it that sometimes the program runs and I get input and >> sometimes it doesn't? My best guess is that the program always runs, [quoted text clipped - 10 lines] > code, but use of those could give the results you describe (avoid > them). That's what I've heard: avoid them like the plague because they aren't reliable and are just messy.
> However if waitFor() returns, that is a string indication that your > program has exited. What is returned by Process.exitValue()? What does > your command line look like? > > /gordon Usually the command is just running the file (could be almost any file) to check for a version number, so it's not a long term situation. In most cases, the command will be run with an argument like "--help" or "/help" to get an output that will give me a unique string somewhere that identifies the specific version. I also know the help info is usually printed to STDERR, so I wasn't checking return codes from exitValue().
It seems that even after waitFor() returns and the process is done, the buffers aren't completely "filled" with all the output yet. There's another post about Thread.join() that I'm looking into.
Thanks for the help!
Hal
Free MagazinesGet these publications absolutely FREE for up to 12 months. There are no hidden fees and no obligation. Simply choose a title, complete the application form and submit it. Read more ...
|
|
|