The hotspot and j2se workspaces both contain a copy of jvm.h.
(These two files have been diverging. That should probably be fixed.)

In both of these files, JVM_Read() is documented like this:

/*
 * Read data from a file decriptor into a char array.
 *
 * fd the file descriptor to read from.
 * buf the buffer where to put the read data.
 * nbytes the number of bytes to read.
 *
 * This function returns -1 on error, and 0 on success.
 */
JNIEXPORT jint JNICALL
JVM_Read(jint fd, char *buf, jint nbytes);


This documentation needs fixing in the following ways:

- This function actually appears to return:
  the number of bytes read, or JVM_IO_ERR, or JVM_IO_INTR

- A maintainer of j2se native code needs guidance regarding the following:

  - When should one use JVM_Read() instead of native read()?

  - How should a caller deal with interrupted system calls?
    Apparently the right thing to do is to use something like this
    (completely untested):


jint JVM_RestartingRead(jint fd, char *buf, jint nbytes)
{
    jint n;
    do { n = JVM_Read(fd, buf, nbytes); }
    while (n == JVM_IO_ERR && errno == EINTR);
    return n;
}

  Perhaps the above function should be made part of the JVM interface?

  - How should a caller deal with JVM_IO_INTR?
    This is extremely tricky, especially given the confusion
    with EINTR.

- "decriptor" typo.
###@###.### 2003-08-11