RFR(trivial): 8231024: Improve the debug info when the output is truncated
ioi.lam at oracle.com
Mon Sep 16 23:20:31 UTC 2019
On 9/16/19 12:44 AM, David Holmes wrote:
> On 16/09/2019 5:09 pm, Jie Fu wrote:
>> Hi David,
>> Sorry for the confusion.
> The confusion is mine, not from you :)
>> On 2019/9/16 下午2:40, David Holmes wrote:
>>> Okay I'm confused. I had assumed that the passed in buffer size must
>>> be different in each case because of the 2014/2032/2050 values in
>>> the message, but in fact they are all sized at 2000.
>> According to the source code, the buffer size can be either
>> 2000(O_BUFLEN)  or 2048(BUFLEN) .
>> So it isn't always 2000.
> It is 2000 is all three examples of the output that you gave.
xmlStream::va_tag() uses 2*K.
>>> That makes no sense
>> I think it makes sense.
>> It is still unclear which one (O_BUFLEN or BUFLEN?) should be
>> increased if the current buffer size wasn't printed.
> You need to know the original caller that passed in the buffer size.
> But we can't just keep increasing the buffer size. We probably need to
> examine the callers to see what amount of information is trying to be
> passed through. Then decide whether to adjust what the callers are
> passing through, or adjust the buffer size - perhaps on a callsite by
> callsite basis rather than just bumping O_BUFLEN or BUFLEN.
I think the message should stop saying "increase O_BUFLEN" -- probably
the reviewers will reject all future increase of O_BUFLEN and will
suggest fixing the caller of outputStream::do_vsnprintf instead.
Callers of outputStream::do_vsnprintf don't necessarily use O_BUFLEN.
E.g., xmlStream::va_tag allocates its own buffer.
VMError::report_and_die() also uses its own buffer -- it uses O_BUFLEN
today, but there's no guarantee or requirement that it always does that.
How about saying something like "outputStream::do_vsnprintf output
truncated -- buffer length is 2000 bytes but 2100 bytes are needed."
>>> - how can we have written e.g. 2014 bytes to a buffer of only 2000
>> The output is truncated since the current buffer size is too small
>> (e.g., 2000 < 2014).
> Ah sorry - I'd missed that aspect of vsnprintf. We're calling:
> 102 int written = os::vsnprintf(buffer, buflen, format, ap);
> but it isn't "written" when it exceeds buflen, it's the required
> buffer size.
I think the variable "written" should be renamed to "required".
>>> Do we have a bug in os::vsnprintf?
>> I don't think so.
>> Thanks a lot.
>> Best regards,
More information about the hotspot-dev