From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from lndn.lancelotsix.com (lndn.lancelotsix.com [51.195.220.111]) by sourceware.org (Postfix) with ESMTPS id 55F893858406 for ; Wed, 29 Sep 2021 21:50:59 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 55F893858406 Received: from Plymouth.lan (unknown [IPv6:2a02:390:9086:0:1f74:c407:abcc:344d]) by lndn.lancelotsix.com (Postfix) with ESMTPSA id B6EFC80D76; Wed, 29 Sep 2021 21:50:57 +0000 (UTC) From: Lancelot SIX To: gdb-patches@sourceware.org Cc: Lancelot SIX , Pedro Alves Subject: [PATCH v4 0/4] Refactor cmd_list_element.var Date: Wed, 29 Sep 2021 22:50:07 +0100 Message-Id: <20210929215011.1489639-1-lsix@lancelotsix.com> X-Mailer: git-send-email 2.33.0 MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Greylist: Sender succeeded SMTP AUTH, not delayed by milter-greylist-4.5.11 (lndn.lancelotsix.com [0.0.0.0]); Wed, 29 Sep 2021 21:50:57 +0000 (UTC) X-Spam-Status: No, score=-4.8 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: gdb-patches@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gdb-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 29 Sep 2021 21:51:02 -0000 Hi, This is a V4 following https://sourceware.org/pipermail/gdb-patches/2021-September/182006.html This version addresses the comments from Pedro. It has also been rebased on top of current trunk. The entire series have been tested on x86_64-linux-gnu without regression observed. All feedbacks welcome. Lancelot. --- Noticable changes since V3: - Patch 1: - setting::get now returns by const reference instead of value. - setting::set has a parameter passed by const ref instead of value. - Since the setting value is returned by ref, remove the setting::get_p function which returned the value by address. The (unique) caller of this function can get the address from the reference instead. - A setting can be costructed from a pre type-erased and pre validated setting::erased_arg value. This is mainly used to avoid turning the best part of add_setshow_cmd_full into a templated function. - Use '= delete' to explicitly disable var_type_uses for any T which is not used to store a setting's value. This improves the compiler error message when such overload is used accidently. - Patch 2: - Use the fact that setting::get returns by reference and avoids making few un necessary copies of strings. - Patch 3: - Follow the changes introduced in patch 1 and allow construction of a setting::erased_arg usign callback functions. - Patch 4: Unchanged. Noticeable changes since V2: - Patch 1: - cmd_list_element.var is now a gdb::optional, and the setting ctor ensures a setting cannot exist if it does not contain a value. - Instead of storing a setting with param_smob (scheme) or parmpy_object (python), keep the var_types and value as they are and create a setting wrapper around them only when required. - Patch 2 is unchanged. - Patch 3: - Store callback function pointers with a type erased 'void (*) ()' instead of using a union. This requires significantly less code. - Add a STORAGE property that can tell the caller if the setting accesses its value via pointer or using the callback methods. This is used in show_remote_protocol_packet_cmd in order to ensure that accessing the setting by pointer is valid. - Patch 4 kept unchanged except for conflict resolutions after rebasing on top of current trunk. Noticeable changes since V1: - Based on the feedbacks form Tom Tromey, the API have been simplified and the amount of templated code have been reduced. Instead of using something like setting.get () in order to access the value of a setting backed by an int (which must be either a var_integer, a var_zinetger or a var_zuinteger_unlimited), we now write setting.get () The 'get' function still includes an assert that setting.var_type is one of the var_types that is backed by an int. In this version, the call site needs to know the mapping between var_types and underlying storage type. - Based on a comment from Simon Marchi, the name of the type that ties together the var_type and the void * pointer has changed from 'param' to 'setting'. - Minor formatting issues have been addressed. --- This patch series proposes to refactor the way cmd_list_element.var is handled. It is meant to be a cleanup series and is intended to offer a basis on which to solve PR/28085. Some commands contain a reference to a variable (setting) that can be set or shown. To do that, the command element contains pieces of information: a void* pointer to global variable, and a var_types field to indicate the type of the setting. The var_type dictates what the data type of variable pointed by the void* pointer must be. When any piece of code within GDB tries to interact with such variable, it first checks the type field and then casts the void* pointer into whatever type pointer it needs to and dereferences it. There are few limitations with this approach I try to address in this patch series: - Any code that interact with a cmd_list_element.var is responsible to do some cast to the appropriate type. This is done in multiple places in the codebase and is quite easy to get wrong. The first patch tries to cover that by introducing a safer abstraction around the void * pointer. The accessors that hide the void * pointer adds runtime checks that the pointer is casted to the appropriate type that matches the var_type of the setting instance. - The source of truth for any setting needs to be the variable pointed to. However, as pointed out by Simon in PR/28085 this is not always true, and there are bugs linked to this. In order to solve this problem, patch 3 allows getter and setter callbacks to be embedded in the abstraction introduced in patch 1. When the setting is accessed, the callback is called to compute or update the value based on the current state of GDB. Patch 2, which was originally proposed by Simon Marchi[1] proposes to change the variable type from char* to std::string for string like setting. Finally Patch 4 does some more refactoring on how we detect and notify observers when a given setting is updated. None of those patches should introduce user visible changes. The entire series have been tested on x86_64 and no regression have been observed on the testsuite. All feedbacks are welcome. Lancelot SIX (3): gdb: Introduce setting construct within cmd_list_element gdb: Have setter and getter callbacks for settings gdb: Setting setter return a bool to tell if the value changed Simon Marchi (1): gdb: make string-like set show commands use std::string variable gdb/auto-load.c | 50 ++-- gdb/breakpoint.c | 22 +- gdb/build-id.c | 4 +- gdb/cli/cli-cmds.c | 148 ++++++----- gdb/cli/cli-decode.c | 459 +++++++++++++++++++++++++++++------ gdb/cli/cli-decode.h | 9 +- gdb/cli/cli-logging.c | 23 +- gdb/cli/cli-option.c | 9 +- gdb/cli/cli-option.h | 4 +- gdb/cli/cli-setshow.c | 188 ++++++-------- gdb/cli/cli-setshow.h | 4 +- gdb/command.h | 335 ++++++++++++++++++++++++- gdb/compile/compile.c | 46 ++-- gdb/corefile.c | 17 +- gdb/defs.h | 4 +- gdb/disasm.c | 11 +- gdb/disasm.h | 2 +- gdb/dwarf2/dwz.c | 2 +- gdb/dwarf2/index-cache.c | 10 +- gdb/dwarf2/read.c | 10 +- gdb/event-top.c | 12 +- gdb/fork-child.c | 7 +- gdb/guile/scm-param.c | 175 +++++++------ gdb/infcmd.c | 14 +- gdb/linux-thread-db.c | 17 +- gdb/main.c | 17 +- gdb/maint-test-options.c | 11 +- gdb/maint-test-settings.c | 8 +- gdb/maint.c | 2 +- gdb/mi/mi-cmd-env.c | 18 +- gdb/proc-api.c | 5 +- gdb/python/py-param.c | 65 +++-- gdb/python/python-internal.h | 2 +- gdb/python/python.c | 29 ++- gdb/remote-sim.c | 7 +- gdb/remote.c | 13 +- gdb/serial.c | 8 +- gdb/solib.c | 20 +- gdb/source.c | 66 ++--- gdb/source.h | 5 +- gdb/stack.c | 22 +- gdb/symfile.c | 49 ++-- gdb/symtab.c | 46 ++-- gdb/target-descriptions.c | 2 +- gdb/top.c | 112 ++++----- gdb/top.h | 2 +- gdb/tracepoint.c | 29 +-- gdb/tracepoint.h | 2 +- 48 files changed, 1395 insertions(+), 727 deletions(-) -- 2.33.0