Remove dependency on apr by using c++11 / c++14 features #255

normanmaurer · 2017-03-23T15:54:31Z

Motivation:

We had some limited use of apr left in our tcnative fork which complicated the deployment. This apr usages can be replaced by using c++.

Modification:

Remove the dependency on apr by using c++11 / c++14 features

Result:

No apr dependeny

normanmaurer · 2017-03-23T15:54:56Z

not 100 % done yet. Will ask for review once I am done

nmittler

@normanmaurer looks great! Just a few nits

nmittler · 2017-03-23T17:40:12Z

openssl-dynamic/src/main/c/ssl.c

    CRYPTO_set_dynlock_create_callback(ssl_dyn_create_function);
    CRYPTO_set_dynlock_lock_callback(ssl_dyn_lock_function);
    CRYPTO_set_dynlock_destroy_callback(ssl_dyn_destroy_function);

-    apr_pool_cleanup_register(p, NULL, ssl_thread_cleanup,
-                              apr_pool_cleanup_null);
+    //apr_pool_cleanup_register(p, NULL, ssl_thread_cleanup,


Looking for a good way to call these again. So stay tuned

nmittler · 2017-03-23T17:40:37Z

openssl-dynamic/src/main/c/ssl.c

-    apr_pool_cleanup_register(tcn_global_pool, NULL,
-                              ssl_init_cleanup,
-                              apr_pool_cleanup_null);
+    //apr_pool_cleanup_register(tcn_global_pool, NULL,


Looking for a good way to call these again. So stay tuned

nmittler · 2017-03-23T17:41:48Z

openssl-dynamic/src/main/c/sslutils.c

 #include "ssl_private.h"
+#include <string.h>


Is this needed? Was it previously brought in by another header?

yep was brought it by another header before.

normanmaurer · 2017-03-23T22:37:44Z

@nmittler @Scottmitch PTAL... it compiles here and on my linux / windows vm. Need to update the g++ version on the ci.

nmittler

Looks great! Goodbye APR! :)

normanmaurer · 2017-03-28T08:22:42Z

@trustin PTAL as well.

Scottmitch · 2017-03-28T15:57:19Z

openssl-dynamic/src/main/c/ssl.c

@@ -1987,3 +1918,8 @@ TCN_IMPLEMENT_CALL(void, SSL, freeX509Chain)(TCN_STDARGS, jlong x509Chain)
    sk_X509_pop_free(chain, X509_free);
 }

+void tcn_ssl_unload() {
+    if (ssl_init_cleanup()) {


should we just have a single method which does both cleanups? the ssl_init_cleanup method is also called in the initialize above, but initialize also calls ssl_thread_setup so it seems like we should cleanup both anyways. It also is awkward to split the cleanup amongst 2 methods in this way.

sure can do

Scottmitch · 2017-03-28T15:58:12Z

openssl-dynamic/src/main/c/ssl_private.h

@@ -212,13 +213,13 @@ struct tcn_ssl_ctxt_t {
    /* TLS ticket key session resumption statistics */

    // The client did not present a ticket and we issued a new one.
-    apr_uint32_t             ticket_keys_new;
+    tcn_atomic_uint32_t             ticket_keys_new;


plz fix the alignment for the variable names so they line up with the other variable names like before this change.

Scottmitch · 2017-03-28T16:00:09Z

openssl-dynamic/src/main/c/tcn_atomic.cpp

+}
+
+void tcn_atomic_uint32_free(tcn_atomic_uint32_t atomic) {
+    auto *p = (std::atomic<uint32_t> *) atomic;


consider removing the temp variable.

also consider passing a ** as a parameter so you can always set *atomic = NULL; in these methods.

this comment applies to all the cpp files.

Scottmitch · 2017-03-28T16:06:26Z

openssl-dynamic/src/main/c/sslcontext.c

-
-    if ((c = apr_pcalloc(p, sizeof(tcn_ssl_ctxt_t))) == NULL) {
-        tcn_ThrowAPRException(e, apr_get_os_error());
+    if ((c = OPENSSL_malloc(sizeof(tcn_ssl_ctxt_t))) == NULL) {


consider just using calloc instead of OPENSSL_malloc. IIUC OpenSSL is not aware of this variable and therefore we don't need to use OpenSSL allocation.

also make sure you use free instead of OPENSSL_free

sounds good... this way we can also remove the memset

Scottmitch · 2017-03-28T16:09:56Z

openssl-dynamic/src/main/c/sslcontext.c

-    return ssl_context_cleanup(c);
+    ssl_context_cleanup(c);
+
+    // TODO: Fix me


can we fix this now ... just make this function void?

like said before I wanted to do the pr without api breakage to make it easier to consume for now.

Doing a followup PR is fine but the return value is not used in Netty. I don't think this will break anything.

Scottmitch · 2017-03-28T16:41:58Z

openssl-dynamic/src/main/native-package/configure.ac

@@ -29,12 +29,14 @@ AC_CANONICAL_HOST
 AC_CANONICAL_SYSTEM

 ${CFLAGS="-O3 -Werror -fno-omit-frame-pointer -Wunused-variable"}
-${CXXFLAGS="-O3 -Werror -fno-omit-frame-pointer -Wunused-variable"}
+# Ensure we support c++11 for atomics etc


c++11 -> c++14

Scottmitch · 2017-03-28T16:44:53Z

openssl-dynamic/src/main/c/tcn_lock_rw.h

+extern "C" {
+#endif // __cplusplus
+
+typedef void* tcn_lock_rw_t;


we have lock_rw, lock_reader, and lock_writer. Consider being consistent/complete with the naming:

lock_reader_writer, lock_reader, and lock_writer

lock_rw, lock_r, and lock_w

Also consider making the file name consistent if you go with the more verbose names.

Scottmitch · 2017-03-28T16:46:24Z

openssl-dynamic/src/main/c/ssl.c

-    }
-
-    value = (struct CRYPTO_dynlock_value *)apr_palloc(p,
+    value = (struct CRYPTO_dynlock_value *)OPENSSL_malloc(


consider declaring and initializing this variable in the same statement.

Scottmitch · 2017-03-28T16:50:36Z

openssl-dynamic/src/main/c/ssl.c

 {
-    UNREFERENCED(data);
    CRYPTO_set_locking_callback(NULL);
    CRYPTO_THREADID_set_callback(NULL);
    CRYPTO_set_dynlock_create_callback(NULL);


it doesn't look like this method takes ownership of the memory [1]. We should take more care to free this memory, and if we set it check if there already exists something and free it.

[1] https://github.com/openssl/openssl/blob/OpenSSL_1_0_2-stable/crypto/cryptlib.c#L376

also be sure to use the ssl_dyn_destroy_function (should be accessible via CRYPTO_get_dynlock_destroy_callback) on the result of CRYPTO_get_dynlock_create_callback.

hmm why would this be needed if we not malloc the callbacks before?

can you clarify? previously a apr_palloc was used to allocate the CRYPTO_dynlock_value structure. This PR uses OPENSSL_malloc instead of the apr_palloc method. Either way there is dynamic memory being allocated. I'm not sure where/if it was freed before, but unless you find that OpenSSL is taking ownership in this situation (just setting the variable to NULL) then we need to ensure this dynamic memory is cleaned up before we lose the reference to it.

Scottmitch · 2017-03-28T16:55:05Z

openssl-dynamic/src/main/c/ssl.c

-    if (rv != APR_SUCCESS) {
-        /* TODO log that fprintf(stderr, "Failed to destroy mutex for dynamic lock %s:%d", l->file, l->line); */
-    }
+    free((void *) l->file);


is the cast necessary? can we just make the type char* in the structure definition if the const is confusing things?

let me see if this works.

normanmaurer · 2017-03-28T17:56:17Z

Doh brain fart ... you are right let me fix it

…

Am 28.03.2017 um 19:52 schrieb Scott Mitchell ***@***.***>: @Scottmitch commented on this pull request. In openssl-dynamic/src/main/c/ssl.c: > { - UNREFERENCED(data); CRYPTO_set_locking_callback(NULL); CRYPTO_THREADID_set_callback(NULL); CRYPTO_set_dynlock_create_callback(NULL); can you clarify? previously a apr_palloc was used to allocate the CRYPTO_dynlock_value structure. This PR uses OPENSSL_malloc instead of the apr_palloc method. Either way there is dynamic memory being allocated. I'm not sure where/if it was freed before, but unless you find that OpenSSL is taking ownership in this situation (just setting the variable to NULL) then we need to ensure this dynamic memory is cleaned up before we lose the reference to it. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or mute the thread.

normanmaurer · 2017-03-28T18:01:33Z

Good point. Let me change it

…

Am 28.03.2017 um 19:55 schrieb Scott Mitchell ***@***.***>: @Scottmitch commented on this pull request. In openssl-dynamic/src/main/c/sslcontext.c: > @@ -317,7 +328,10 @@ TCN_IMPLEMENT_CALL(jint, SSLContext, free)(TCN_STDARGS, jlong ctx) UNREFERENCED_STDARGS; TCN_ASSERT(ctx != 0); - return ssl_context_cleanup(c); + ssl_context_cleanup(c); + + // TODO: Fix me Doing a followup PR is fine but the return value is not used in Netty. I don't think this will break anything. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or mute the thread.

Scottmitch · 2017-03-28T21:26:46Z

openssl-dynamic/src/main/c/ssl.c

-        /* TODO log that fprintf(stderr, "Failed to create subpool for dynamic lock"); */
-        return NULL;
-    }
+    value = (struct CRYPTO_dynlock_value *)malloc(sizeof(struct CRYPTO_dynlock_value));


#255 (comment)

consider declaring and initializing this variable in the same statement.

Scottmitch · 2017-03-28T21:29:59Z

openssl-dynamic/src/main/c/ssl.c

+
+    CRYPTO_set_locking_callback(NULL);
+    CRYPTO_THREADID_set_callback(NULL);
+    CRYPTO_set_dynlock_create_callback(NULL);


#255 (comment)

Can you provide a link as to where the previous callback is freed if we just set the callback to NULL? When I looked at OpenSSL src it seemed like this created a memory leak.

it is not yet, will do today (and it was not freed before in tcnative as well)

Actually I think all is good now. Please ping me once you are awake and we can talk about it in more detail if you have doubts. Maybe I am just missing something but from my look into the source etc its just fine now

I didn't track it all the way down but found that there are links to the callbacks retained in the engines ... so ENGINE_cleanup maybe the place where these callbacks are used to clean stuff up. my concern was we were missing a call which notifies OpenSSL to use these callbacks. @davidben - can you shed some light on this situation?

Scottmitch · 2017-03-28T22:10:47Z

openssl-dynamic/src/main/c/ssl.c

 }
+#endif

 TCN_IMPLEMENT_CALL(jint, SSL, initialize)(TCN_STDARGS, jstring engine)


#255 (comment)

it looks like this can just be void because we throw an exception upon failure anyways.

looks like the method return type was changed in the java header, but not here ... seems like there will be a mismatch

yeah... noe done yet.

Scottmitch · 2017-03-28T22:14:55Z

openssl-dynamic/src/main/c/ssl.c

@@ -2113,3 +2043,7 @@ TCN_IMPLEMENT_CALL(jbyteArray, SSL, getOcspResponse)(TCN_STDARGS, jlong ssl) {
    return value;
 #endif
 }
+
+void tcn_ssl_unload() {
+    ssl_init_cleanup();


can you combine tcn_ssl_unload and ssl_init_cleanup? seems unnecessary to have a function which only calls a single function?

Scottmitch · 2017-03-28T22:16:54Z

openssl-dynamic/src/main/c/sslcontext.c

@@ -87,14 +84,19 @@ static apr_status_t ssl_context_cleanup(void *data)
            free(c->password);
            c->password = NULL;
        }
+
+        tcn_atomic_uint32_free(&(c->ticket_keys_new));


&(c->ticket_keys_new) -> &c->ticket_keys_new ? The parens shouldn't be necessary, right?

Scottmitch · 2017-03-28T22:28:19Z

openssl-dynamic/src/main/c/tcn_lock_rw.cpp

+
+tcn_lock_rw_t tcn_lock_rw_new() {
+    // Once we switch to c++17 we should use std::shared_mutex
+    return (tcn_lock_rw_t) new std::shared_timed_mutex();


IIRC in C++ it is ambiguous if you try to use () for the no argument constructor without the new keyword ... new may make the situation unambiguous but consider dropping the () which may be more C++ish.

http://stackoverflow.com/a/2318731

Scottmitch · 2017-03-28T22:29:49Z

openssl-dynamic/src/main/c/tcn_lock_rw.cpp

+
+tcn_lock_w_t tcn_lock_w_acquire(tcn_lock_rw_t lock) {
+    auto *m = (std::shared_timed_mutex *) lock;
+    return (tcn_lock_w_t) new std::lock_guard<std::shared_timed_mutex>(*m);


#255 (comment)

why not std::unique_lock?

also consider removing the temp variable so you don't have to worry about auto or types:

return (tcn_lock_w_t) new std::lock_guard<std::shared_timed_mutex>(*((std::shared_timed_mutex*) lock));

Scottmitch · 2017-03-28T22:33:32Z

openssl-dynamic/src/main/c/tcn_lock_rw.cpp

+}
+
+tcn_lock_r_t tcn_lock_r_acquire(tcn_lock_rw_t lock) {
+    auto *m = (std::shared_timed_mutex *) lock;


consider removing the temp variable so you don't have to worry about auto or types:

return (tcn_lock_r_t) new std::shared_lock<std::shared_timed_mutex>(*((std::shared_timed_mutex*) lock));

Scottmitch · 2017-03-28T22:35:54Z

openssl-dynamic/src/main/c/tcn_thread.h

+#ifndef TCN_THREAD_H
+#define TCN_THREAD_H
+
+#include <stddef.h>


you may want to be sure you include the cpp headers if you are in cpp, and the c headers if you are in c. They may define types differently or have different behaviors.

#ifdef __cplusplus #include <cstddef> extern "C" { #else #include <stddef.h> #endif // __cplusplus

Scottmitch · 2017-03-28T22:40:30Z

openssl-dynamic/src/main/c/tcn_atomic.h

+#ifndef TCN_ATOMIC_H
+#define TCN_ATOMIC_H
+
+#include <stdint.h>


you may want to be sure you include the cpp headers if you are in cpp, and the c headers if you are in c. They may define types differently or have different behaviors.

#ifdef __cplusplus #include <cstdint> extern "C" { #else #include <stdint.h> #endif // __cplusplus

Scottmitch

some small cleanup ... and just the memory management question left then lgtm.

Scottmitch · 2017-03-29T17:38:23Z

openssl-dynamic/src/main/c/ssl.c

 /*
 * Dynamic lock creation callback
 */
 static struct CRYPTO_dynlock_value *ssl_dyn_create_function(const char *file,
                                                     int line)
 {
    struct CRYPTO_dynlock_value *value;
-    apr_pool_t *p;
-    apr_status_t rv;
+    value = (struct CRYPTO_dynlock_value *)malloc(sizeof(struct CRYPTO_dynlock_value));


#255 (comment)

consider declaring and initializing this variable in the same statement.

struct CRYPTO_dynlock_value* value = (struct CRYPTO_dynlock_value*) malloc(sizeof(struct CRYPTO_dynlock_value));

Scottmitch · 2017-03-29T17:39:27Z

openssl-dynamic/src/main/c/ssl.c

@@ -106,7 +99,7 @@ struct TCN_bio_bytebuffer {
 * so that the return value for SSL_OP_FUTURE_WORKAROUND will only be
 * reported by versions that actually support that specific workaround.
 */
-static const jint supported_ssl_opts = 0
+static const jlong supported_ssl_opts = 0


lets hold off on this change and just rebase when #258 is merged

Scottmitch · 2017-03-29T18:17:17Z

openssl-dynamic/src/main/c/ssl.c

+
+    CRYPTO_set_locking_callback(NULL);
+    CRYPTO_THREADID_set_callback(NULL);
+    CRYPTO_set_dynlock_create_callback(NULL);


I didn't track it all the way down but found that there are links to the callbacks retained in the engines ... so ENGINE_cleanup maybe the place where these callbacks are used to clean stuff up. my concern was we were missing a call which notifies OpenSSL to use these callbacks. @davidben - can you shed some light on this situation?

Scottmitch · 2017-03-29T18:19:00Z

openssl-dynamic/src/main/c/sslcontext.c

@@ -74,7 +71,7 @@ static apr_status_t ssl_context_cleanup(void *data)
        }
        c->alpn_proto_len = 0;

-        apr_thread_rwlock_destroy(c->mutex);
+        tcn_lock_rw_free(&(c->ticket_keys_lock));


&(c->ticket_keys_lock) -> &c->ticket_keys_lock

Scottmitch · 2017-03-29T18:19:20Z

openssl-dynamic/src/main/c/ssl.c

@@ -2113,3 +2041,4 @@ TCN_IMPLEMENT_CALL(jbyteArray, SSL, getOcspResponse)(TCN_STDARGS, jlong ssl) {
    return value;
 #endif
 }
+


kill the extra new line

trustin

Way to go!

trustin · 2017-03-31T02:43:10Z

openssl-dynamic/src/main/c/ssl.c

    }

 #if OPENSSL_VERSION_NUMBER < 0x10100000L
    if (SSLeay() < 0x0090700L) {
        TCN_FREE_CSTRING(engine);
-        tcn_ThrowAPRException(e, APR_EINVAL);
+        tcn_ThrowException(e, "openssl version too old");


How about appending the version number and the minimum required version number?

trustin · 2017-03-31T02:49:21Z

openssl-dynamic/src/main/native-package/configure.ac

@@ -29,12 +29,14 @@ AC_CANONICAL_HOST
 AC_CANONICAL_SYSTEM

 ${CFLAGS="-O3 -Werror -fno-omit-frame-pointer -Wunused-variable"}
-${CXXFLAGS="-O3 -Werror -fno-omit-frame-pointer -Wunused-variable"}
+# Ensure we support c++14 for atomics etc
+${CXXFLAGS="-O3 -Werror -fno-omit-frame-pointer -Wunused-variable -std=c++14"}


Perhaps should we static-link libgcc and libstdc++? -static-libgcc -static-libstdc++
Also, we may eventually want to link against musl or uClibc instead of GLIBC so our binary works on more distros. http://stackoverflow.com/a/26306630

I think we not want to do any static linking in our dynamic build. Not sure about the "-static" ones tho. @Scottmitch @nmittler WDYT ?

I'm not sure if the situation has changed, but I had an impression that libstdc++ has worse backward ABI compatibility historically. Maybe I'm completely wrong, though.

Lets handle this as a followup PR if it becomes an issue or is known to improve things.

Scottmitch

few more comments then lgtm

Scottmitch · 2017-03-31T17:51:01Z

openssl-dynamic/src/main/c/ssl.c


 #ifndef OPENSSL_NO_ENGINE
    if (J2S(engine)) {
        ENGINE *ee = NULL;
-        apr_status_t err = APR_SUCCESS;
+        int err = 0;


just use bool?

Scottmitch · 2017-03-31T17:52:41Z

openssl-dynamic/src/main/c/ssl.c

@@ -721,19 +656,11 @@ TCN_IMPLEMENT_CALL(jint, SSL, initialize)(TCN_STDARGS, jstring engine)
    if (r) {


(not introduce by this PR, but while we are here) r is an integer type ... lets use it like an integer: r != 0

Scottmitch · 2017-03-31T17:53:45Z

openssl-dynamic/src/main/c/tcn_atomic.h

+#ifndef TCN_ATOMIC_H
+#define TCN_ATOMIC_H
+
+


nit: kill line

Scottmitch · 2017-03-31T17:54:33Z

openssl-dynamic/src/main/c/tcn_thread.h

+#ifndef TCN_THREAD_H
+#define TCN_THREAD_H
+
+


nit: kill line

Scottmitch · 2017-05-06T22:31:41Z

@normanmaurer - are you planning on pulling this in?

normanmaurer · 2017-05-07T07:07:22Z

Yes just wanted to do a few more tests

…

Am 07.05.2017 um 00:31 schrieb Scott Mitchell ***@***.***>: @normanmaurer - are you planning on pulling this in? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

Motivation: We had some limited use of apr left in our tcnative fork which complicated the deployment. This apr usages can be replaced by using c++. Modification: Remove the dependency on apr by using c++11 / c++14 features Result: No apr dependeny

normanmaurer · 2017-07-18T12:34:34Z

@Scottmitch PTAL again.. rebased on latest master and should are ready to get pulled in.

normanmaurer · 2017-07-18T20:05:53Z

After @Scottmitch mentioned to me that centos 6 glibc may not support c++14 I checked in more detail and found out it even not support c++11 . So I guess we will need to life with apr ...

@Scottmitch WDYT ?

Scottmitch · 2017-07-18T17:32:33Z

openssl-dynamic/src/main/c/ssl.c

-    ssl_lock_cs = apr_palloc(p, ssl_lock_num_locks * sizeof(*ssl_lock_cs));
+    if (!ssl_initialized)
+        return;
+    ssl_initialized = 0;


do we need to consider concurrency/visibility for ssl_initialized? can we just remove ssl_initialized and consider it a programing error if initialized is called multiple times (can this happen in Netty)?

Scottmitch · 2017-07-18T22:48:22Z

openssl-dynamic/src/main/c/sslcontext.c

        goto cleanup;
    }

    c->protocol = protocol;
    c->mode     = mode;
    c->ctx      = ctx;
-    c->pool     = p;
+    c->ticket_keys = tcn_atomic_uint32_new();


is this correct? ticket_keys is of type tcn_ssl_ticket_key_t but this is assigning it to a new atomic uint32 of type void*, right?

Was the intention that this should be ticket_keys_new and ticket_keys should be initialized in some other way? It is freed above using OPENSSL_free.

Scottmitch · 2017-07-19T15:31:58Z

@normanmaurer - I would like to still remove APR ... we should investigate building a glibc shared library, either publish these instructions or the library, and have folks on older distributions depend upon this instead of apr. Newer distributions which have glibc that support c++14 (eventually every one over time) will require not additional dependency. The statically compiled artifacts may be a bit more tricky due to licensing of glibc.

kvc-code · 2017-09-18T21:38:53Z

How does this affect the ability to build against musl C library on Alpine? At present we at least have the ability to build for ourselves on Alpine but sounds like this change may lock that out as well.

normanmaurer · 2019-09-23T12:56:15Z

Let me close this for now

nmittler approved these changes Mar 23, 2017

View reviewed changes

normanmaurer force-pushed the remove_apr branch from be9aace to 105f955 Compare March 23, 2017 22:29

normanmaurer force-pushed the remove_apr branch 3 times, most recently from d242b52 to 946520b Compare March 24, 2017 00:28

nmittler approved these changes Mar 24, 2017

View reviewed changes

normanmaurer requested review from trustin and Scottmitch March 28, 2017 08:22

normanmaurer self-assigned this Mar 28, 2017

Scottmitch requested changes Mar 28, 2017

View reviewed changes

normanmaurer force-pushed the remove_apr branch from 946520b to 06c4109 Compare March 28, 2017 19:16

Scottmitch requested changes Mar 28, 2017

View reviewed changes

normanmaurer force-pushed the remove_apr branch from dd5b0f7 to 8960e1e Compare March 29, 2017 06:28

Scottmitch requested changes Mar 29, 2017

View reviewed changes

normanmaurer force-pushed the remove_apr branch 3 times, most recently from 56646cd to 21c6ffb Compare March 29, 2017 20:29

normanmaurer added this to the 2.0.1.Final milestone Mar 29, 2017

normanmaurer added cleanup improvement labels Mar 29, 2017

trustin requested changes Mar 31, 2017

View reviewed changes

Scottmitch requested changes Mar 31, 2017

View reviewed changes

normanmaurer force-pushed the remove_apr branch from 21c6ffb to 534eb6c Compare May 27, 2017 11:44

normanmaurer modified the milestones: 2.0.4.Final, 2.0.5.Final, 2.0.6.Final Jun 23, 2017

normanmaurer force-pushed the remove_apr branch from 6ba710c to 1c4e1c5 Compare July 18, 2017 12:18

normanmaurer closed this Jul 18, 2017

normanmaurer reopened this Jul 18, 2017

Scottmitch requested changes Jul 18, 2017

View reviewed changes

normanmaurer modified the milestones: 2.0.6.Final, 2.0.7.Final Aug 18, 2017

stresler mentioned this pull request Sep 14, 2017

Unable to use pre-built binaries on alpine linux #111

Closed

normanmaurer modified the milestones: 2.0.7.Final, 2.0.8.Final Nov 3, 2017

normanmaurer modified the milestones: 2.0.8.Final, 2.0.9.Final Apr 13, 2018

normanmaurer force-pushed the master branch 2 times, most recently from 0d59905 to 4937622 Compare May 31, 2018 06:51

normanmaurer modified the milestones: 2.0.9.Final, 2.0.11.Final Jun 18, 2018

normanmaurer removed this from the 2.0.11.Final milestone Mar 22, 2019

normanmaurer closed this Sep 23, 2019

		@@ -721,19 +656,11 @@ TCN_IMPLEMENT_CALL(jint, SSL, initialize)(TCN_STDARGS, jstring engine)
		if (r) {

Remove dependency on apr by using c++11 / c++14 features #255

Remove dependency on apr by using c++11 / c++14 features #255

Conversation

normanmaurer commented Mar 23, 2017

normanmaurer commented Mar 23, 2017

nmittler left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

normanmaurer commented Mar 23, 2017

nmittler left a comment

Choose a reason for hiding this comment

normanmaurer commented Mar 28, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

normanmaurer commented Mar 28, 2017 via email

normanmaurer commented Mar 28, 2017 via email

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Scottmitch left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

trustin left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

trustin Mar 31, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Scottmitch Mar 31, 2017 • edited Loading

Choose a reason for hiding this comment

Scottmitch left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

trustin Mar 31, 2017 •

edited

Loading

Scottmitch Mar 31, 2017 •

edited

Loading