improve mutation_data write_to/read_from #499
Conversation
| writer.write(update.data.data(), update.data.length()); | ||
| writer.flush(); | ||
|
|
||
| message_ex* msg = (message_ex*)to; |
There was a problem hiding this comment.
message_ex is not exposed to upper apps and frameworks, which only invokes the core features through C api (and upper wrappers).
There was a problem hiding this comment.
It there any solution about it? I think the optimization is meaningful.
On Fri, Jul 29, 2016 at 2:17 PM, Zhenyu Guo [email protected]
wrote:
In src/dist/replication/lib/mutation.cpp
#499 (comment):{
writer.write(update.data.data(), update.data.length());writer.flush();
message_ex\* msg = (message_ex*)to;message_ex is not exposed to upper apps and frameworks, which invokes the
core features through C api eventually.—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/imzhenyu/rDSN/pull/499/files/4ed36f13e7f75d282315ba8284d1003944ed1b93#r72745993,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAgQPZD6Maf_qEDZxotGoJcPMwibGYVWks5qaZsPgaJpZM4JX4Dq
.
There was a problem hiding this comment.
you may add a new dsn_msg_write_append C api which requires the callers to manage the buffer life-time correctly.
qinzuoyan
changed the title
|
@imzhenyu , I have added a c-api of "dsn_msg_write_append", please review it. |
| @@ -785,6 +785,9 @@ extern DSN_API void dsn_msg_write_next( | |||
| /*! commit the write buffer after the message content is written with the real written size */ | |||
| extern DSN_API void dsn_msg_write_commit(dsn_message_t msg, size_t size); | |||
|
|
|||
| /*! append a buffer to the message, the "data" can be reinterpreted cast to (blob*) */ | |||
| extern DSN_API void dsn_msg_write_append(dsn_message_t msg, dsn_blob_t data); | |||
There was a problem hiding this comment.
dsn_blob_t is not a POD or C structure, therefore we cannot put it in this C API. May use (dsn_message_t, void*, size_t) instead.
There was a problem hiding this comment.
but then we lost the holder of blob, or just copy data?
On Fri, Jul 29, 2016 at 4:08 PM, Zhenyu Guo [email protected]
wrote:
In include/dsn/c/api_layer1.h
#499 (comment):@@ -785,6 +785,9 @@ extern DSN_API void dsn_msg_write_next(
/*! commit the write buffer after the message content is written with the real written size */
extern DSN_API void dsn_msg_write_commit(dsn_message_t msg, size_t size);+/! append a buffer to the message, the "data" can be reinterpreted cast to (blob) */
+extern DSN_API void dsn_msg_write_append(dsn_message_t msg, dsn_blob_t data);dsn_blob_t is not a POD or C structure, therefore we cannot put it in this
C API. May use (dsn_message_t, void*, size_t) instead.—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/imzhenyu/rDSN/pull/499/files/ea230134983ab865c8bb7b6e2110776c7d14b9ac#r72755516,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAgQPejEDsHaGiYDGDrmUtjSEcIL9P7Dks5qabUTgaJpZM4JX4Dq
.
There was a problem hiding this comment.
The caller needs to guarantee the life-time; or we have to copy data.
There was a problem hiding this comment.
The holder is a shared_ptr, so the life-time seems unable to handle.... can
you think out some resolution to avoid memory copy?
On Fri, Jul 29, 2016 at 4:44 PM, Zhenyu Guo [email protected]
wrote:
In include/dsn/c/api_layer1.h
#499 (comment):@@ -785,6 +785,9 @@ extern DSN_API void dsn_msg_write_next(
/*! commit the write buffer after the message content is written with the real written size */
extern DSN_API void dsn_msg_write_commit(dsn_message_t msg, size_t size);+/! append a buffer to the message, the "data" can be reinterpreted cast to (blob) */
+extern DSN_API void dsn_msg_write_append(dsn_message_t msg, dsn_blob_t data);The caller needs to handle the life-time explicitly; or we have to copy
data.—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/imzhenyu/rDSN/pull/499/files/ea230134983ab865c8bb7b6e2110776c7d14b9ac#r72759513,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAgQPXyekSSBRkBBDcDa5ne71m3UAcGTks5qab1wgaJpZM4JX4Dq
.
There was a problem hiding this comment.
This is what is ugly about C interface, unfortunately. That is why in certain cases I would prefere ref_counter or instrusive ptr instead of shared-ptr:-( I'll think about this though to see how we can enable this more elegantly.
There was a problem hiding this comment.
well, now I will discard the optimization and leave it to future.
On Fri, Jul 29, 2016 at 5:31 PM, Zhenyu Guo [email protected]
wrote:
In include/dsn/c/api_layer1.h
#499 (comment):@@ -785,6 +785,9 @@ extern DSN_API void dsn_msg_write_next(
/*! commit the write buffer after the message content is written with the real written size */
extern DSN_API void dsn_msg_write_commit(dsn_message_t msg, size_t size);+/! append a buffer to the message, the "data" can be reinterpreted cast to (blob) */
+extern DSN_API void dsn_msg_write_append(dsn_message_t msg, dsn_blob_t data);This is what is ugly about C interface, unfortunately. That is why in
certain cases I would prefere ref_counter or instrusive ptr instead of
shared-ptr:-( I'll think about this though to see how we can enable this
more elegantly.—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/imzhenyu/rDSN/pull/499/files/ea230134983ab865c8bb7b6e2110776c7d14b9ac#r72765645,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAgQPbwWAMkLxrIkYmRDBcyBPWZX73lbks5qach4gaJpZM4JX4Dq
.
|
we leave optimization of "directly append blob to message_ex to avoid unnecessary memory copy" to future, refer to #501 |
|
@imzhenyu , is it ok now? |
improvement: