1748 replies to this topic

[zakalibit]

thanks, filtering part will follow, I think you going to like the solution I came up with

[Erik Slagter]

I have one remark on the code, the string "Transfer-Encoding: chunked" is compared case-dependent. AFAIK http doesn't require this specific capitalisation, so IMHO it's better to do a case-independent compare, changing strncmp into strncasecmp should suffice.

 

Also I see a few one-liners

Edited by Erik Slagter, 20 November 2013 - 19:14.

[zakalibit]

I have one remark on the code, the string "Transfer-Encoding: chunked" is compared case-dependent. AFAIK http doesn't require this specific capitalisation, so IMHO it's better to do a case-independent compare, changing strncmp into strncasecmp should suffice.

 

Also I see a few one-liners

Have not seen yet a problem with that, but agree with your point. 

[zakalibit]

You could say the same about other string comparisons.

Those one liners makes at least for me the code more readable and what about the existing one liners? I think is pointless talking about one liners, they look just fine.

Edited by zakalibit, 20 November 2013 - 20:34.

[zakalibit]

Here you go, packet boundary aligned reads

https://sourceforge....ge-requests/28/

[pieterg]

applied, but changed it to use iTsSource::packetSize, instead of fixed 188

[zakalibit]

applied, but changed it to use iTsSource::packetSize, instead of fixed 188

cool, that is  good idea, did not know about that variable

[zakalibit]

line 253  needs to change:

 

- if (partialPktSz > 0 && partialPktSz < sizeof(partialPkt))

+ if (partialPktSz > 0 && partialPktSz < packetSize)

 

because you have increased the buffer size to 192 and we need to save only incomplete packets.

I do not understand that value, the size can be 188 or 204 if correction bytes are added to the packet

Edited by zakalibit, 21 November 2013 - 00:04.

[pieterg]

yes, you're right I missed that.

192 = 188 + external timestamp (for timekeeping in scrambled data, used by m2ts)

[athoik]

Hello,


When we are using TS streams (1:0:1) the service.streamed() is always null. So Converter IsStream on ServiceInfo returns also false.

The following patch fixes streamed function to return correct information when we are using TS stream, based on m_is_stream flag.

--- a/lib/service/servicedvb.cpp
+++ b/lib/service/servicedvb.cpp
@@ -1755,6 +1755,29 @@ RESULT eDVBServicePlay::rdsDecoder(ePtr<iRdsDecoder> &ptr)
 	return 0;
 }
 
+RESULT eDVBServicePlay::streamed(ePtr<iStreamedService> &ptr)
+{
+	if (m_is_stream)
+	{
+		ptr = this;
+		return 0;
+	}
+	ptr = 0;
+	return -1;
+}
+
+ePtr<iStreamBufferInfo> eDVBServicePlay::getBufferCharge()
+{
+	/** FIXME **/
+	return 0;
+}
+
+int eDVBServicePlay::setBufferSize(int size)
+{
+	/** FIXME **/
+        return 0;
+}
+
 RESULT eDVBServicePlay::getName(std::string &name)
 {
 	if (m_is_pvr)
diff --git a/lib/service/servicedvb.h b/lib/service/servicedvb.h
index 038c978..1c2f113 100644
--- a/lib/service/servicedvb.h
+++ b/lib/service/servicedvb.h
@@ -86,7 +86,8 @@ class eDVBServicePlay: public eDVBServiceBase,
 		public iAudioTrackSelection, public iAudioChannelSelection,
 		public iSubserviceList, public iTimeshiftService,
 		public iCueSheet, public iSubtitleOutput, public iAudioDelay,
-		public iRdsDecoder, public iStreamableService
+		public iRdsDecoder, public iStreamableService, 
+		public iStreamedService
 {
 	DECLARE_REF(eDVBServicePlay);
 public:
@@ -111,7 +112,12 @@ public:
 	RESULT audioDelay(ePtr<iAudioDelay> &ptr);
 	RESULT rdsDecoder(ePtr<iRdsDecoder> &ptr);
 	RESULT keys(ePtr<iServiceKeys> &ptr) { ptr = 0; return -1; }
-	RESULT streamed(ePtr<iStreamedService> &ptr) { ptr = 0; return -1; }
+
+		// iStreamedService
+	RESULT streamed(ePtr<iStreamedService> &ptr);
+	ePtr<iStreamBufferInfo> getBufferCharge();
+	int setBufferSize(int size);
+
 
 		// iPauseableService
 	RESULT pause();

 0001-Implement-streamed-call-must-return-not-null-value-i.patch.txt   2.27KB   1 downloads

Thanks/Ευχαριστώ

[pieterg]

thx, applied (minus the whitespace error)

[athoik]

Hello,

The following patch will enable a867 and af9035 for vu 3.1 kernel.

--- a/meta-openpli/recipes-linux/linux/linux-vuplus-3.1.1.inc
+++ b/meta-openpli/recipes-linux/linux/linux-vuplus-3.1.1.inc
@@ -3,7 +3,7 @@ LICENSE = "GPL"
 SECTION = "kernel"
 KV = "3.1.1"
 
-PR_INC = ".8"
+PR_INC = ".9"
 
 SRCREV = "r2"
 
@@ -22,6 +22,8 @@ SRC_URI += "http://archive.vuplus.com/download/kernel/linux-${KV}_${SRCREV}.tar.
         file://em28xx_fix_terratec_entries.patch \
         file://em28xx_add_terratec_h5_rev3.patch \
 	file://fix-dvb-siano-sms-order.patch \
+	file://dvb-usb-af9035.patch \
+	file://dvb-usb-a867.patch \
         file://${MACHINE}_defconfig \
         "

Patches where available but they didn't apply nicely on 3.1 kernel (Makefile patching was failing).

I don't have such a usb stick in order to test it. Compiled kernel modules are available here : http://openpli.org/f...not-on-vu-solo/ , altough the reporter never downloaded modules (probably black hole suck him )

Attached Files

•  0001-vuplus-3.1.1-Add-missing-patches-for-a867-and-af9035.zip   274.69KB   5 downloads

Edited by athoik, 1 December 2013 - 12:46.

[athoik]

Hello,

Can you please commit above?

(because i am preparing new patch for siano dvb-t sticks and PR_INC will change from 9 to 10, actually every kernel will increase PR_INC see here http://openpli.org/f...er/#entry389987)

[athoik]

Hello,
 
Here is a patch to handle deadlock when inserting EPG for IPTV. More information here: http://openpli.org/f...els-in-general/
 
 

When we are destroying eHttpStream object and connectionStatus is
busy, we have deadlock at thread kill. We need to call thread kill
when connectionStatus changes to non busy state.

diff --git a/lib/base/httpstream.cpp b/lib/base/httpstream.cpp
index 1aeaf3f..b18025b 100644
--- a/lib/base/httpstream.cpp
+++ b/lib/base/httpstream.cpp
@@ -1,5 +1,6 @@
 #include <cstdio>
 #include <openssl/evp.h>
+#include <unistd.h> // for usleep

 #include <lib/base/httpstream.h>
 #include <lib/base/eerror.h>
@@ -21,6 +22,11 @@ eHttpStream::eHttpStream()
 eHttpStream::~eHttpStream()
 {
        free(tmpBuf);
+       while (connectionStatus == BUSY)
+       {
+               eDebug("~eHttpStream:: Connection is busy, wait 200 msec...");
+               usleep(200000);  // wait 200 msec
+       }
        kill(true);
        close();
 }

 
 0001-httpstream-Avoid-deadlock-on-eHttpStream-destructor.patch.txt   1004bytes   3 downloads
 
Thanks.

Edited by athoik, 4 December 2013 - 19:30.

[littlesat]

In fact this is not the way how deadlocks should be avoided....

[athoik]

But when Connect is in progress (BUSY) and we are (kill and then) calling pthread_cancel and then pthread_join the thread deadlocks in join.

The destructor must inform Connect to "disconnect" before pthread_join.

But how this is possible?
Send a signal (eg SIGUSR1, but i cannot see any signal handler inside Connect in wrappers.cpp)?

Thanks.

[pieterg]

Doesn't this race? (can the connection change it's state to BUSY again?)

[athoik]

It can be FAILED or CONNECTED after BUSY.

http://sourceforge.n...stream.cpp#l190

BUSY is only when thread starts.

[pieterg]

so where does it hang? Doesn't the thread eventually return, even though the connect might not be aborted right away?

[athoik]

See:

eHttpStream::Start thread
~eHttpStream: calling kill
~eHttpStream: calling close
send cancel to thread
thread: before openUrl
openUrl: before Connect
cancel received by thread
thread join
thread joined 0
eHttpStream::Start thread
thread: before openUrl
openUrl: before Connect
~eHttpStream: calling kill
~eHttpStream: calling close
send cancel to thread
cancel received by thread
thread join

Thread join blocks if we call kill after OpenUrl.