Replace MEM_DREAD_CTRL pool with MEMPROXY class pool #1924
Conversation
... to a MEMPROXY class pool. Replacing memory allocate and destruct with C++ new/delete and in-class initialization.
rousskov
changed the title
There was a problem hiding this comment.
I adjusted PR title/description to match recent commit ed1a732.
Before advancing this new PR (including fixing build tests) and requesting my re-review, please cooperate to merge PRs awaiting your review for a long time, including these three:
- Upgrade generated parse_line() to accept constant SBuf #1840 opened Jun 11, 2024: "Upgrade generated parse_line() to accept constant SBuf"
- Bug 5363: Handle IP-based X.509 SANs better #1793 opened Apr 25, 2024: "Bug 5363: Handle IP-based X.509 SANs better"
- Bug 5383: handleNegotiationResult() level-2 debugs() crash #1856 opened Jun 30, 2024: "Bug 5383: handleNegotiationResult() level-2 debugs() crash"
| DRCB *handler = {}; | ||
| void *client_data = {}; |
There was a problem hiding this comment.
By default, please use nullptr to initialize raw pointers.
| DRCB *handler = {}; | |
| void *client_data = {}; | |
| DRCB *handler = nullptr; | |
| void *client_data = nullptr; |
Alternatively, do not initialize data members that are supposed to be set by the class constructor. This approach allows the compiler to warn us if we forget to copy constructor parameter to a class data member.
| ctrl_dat->end_of_file = 0; | ||
| ctrl_dat->handler = handler; | ||
| ctrl_dat->client_data = cbdataReference(client_data); | ||
| const auto ctrl_dat = new dread_ctrl(fd, offset, req_len, buf, handler, cbdataReference(client_data)); |
There was a problem hiding this comment.
I recommend moving cbdataReference() call inside dread_ctrl constructor and placing the matching cbdataReferenceDone() call inside its destructor. Doing so will lower several red flags:
- Official diskHandleRead() code sometimes does not call cbdataReferenceDone() when deleting ctrl_dat object.
- PR code prohibits dread_ctrl object moving or copying, but there is no obvious reason to do that (now) because the class effectively remains a POD.
I do not insist on this change despite recommending it.
rousskov
added
the
S-waiting-for-author
Replacing memory allocate and destruct with C++ new/delete
and in-class initialization.