Hello, New bugfix for PR19377 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=19377). This is basically an extension of what I did before for PR17314 except it also fixes this other bug.
I hope I didn't over-comment in the code ... better to say too much than too little! It's a niche bug so I thought it could do with a little explanation. Added 1 new regression test. Bootstraps fine on x86_64-pc-linux-gnu for x86_64-pc-linux-gnu. Hopefully no formatting problems; checked with git gcc-verify and check_GNU_style.sh. ----------- REGRESSION ANALYSIS ----------- No regressions reported. I only ran the c++ regression tests since this is a c++ front-end diagnostics code bug (i.e. it should have no effect on compilation, or any other languages). G++ (CLEAN) RESULTS # of expected passes 203705 # of unexpected failures 2 # of expected failures 989 # of unsupported tests 8714 G++ (PATCHED) RESULTS # of expected passes 203717 # of unexpected failures 2 # of expected failures 989 # of unsupported tests 8714 The extra passes are from my new regression test. Let me know if there are any issues. Kind regards, Anthony Sharp
From e064f8d010baee288c47cce1981be80515101f0d Mon Sep 17 00:00:00 2001 From: Anthony Sharp <anthonyshar...@gmail.com> Date: Thu, 4 Feb 2021 12:01:59 +0000 Subject: [PATCH] c++: Check for using decl in private parent access [PR19377] This bug was already mostly fixed by the patch for PR17314. This patch continues that by ensuring that where a using decl is used, causing an access failure to a child class because the using decl is private, the compiler correctly points to the using decl as the source of the problem. Checks for the use of using decls in a parent that had private access to decl (but the child had no access) in order to ascertain the best diagnostic decl location. gcc/cp/ChangeLog: * semantics.c (get_class_access_diagnostic_decl): New function. (enforce_access): Altered function. gcc/testsuite/ChangeLog: * g++.dg/pr19377.C: New test. --- gcc/cp/semantics.c | 93 +++++++++++++++++++++++++++------- gcc/testsuite/g++.dg/pr19377.C | 21 ++++++++ 2 files changed, 95 insertions(+), 19 deletions(-) create mode 100644 gcc/testsuite/g++.dg/pr19377.C diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c index 73834467fca..6d4ef683d65 100644 --- a/gcc/cp/semantics.c +++ b/gcc/cp/semantics.c @@ -256,6 +256,62 @@ pop_to_parent_deferring_access_checks (void) } } +/* Called from enforce_access. A class has attempted (but failed) to access + DECL. It is already established that a baseclass of that class, + PARENT_BINFO, has private access to DECL. Examine certain special cases to + generate a diagnostic decl location. If no special cases are found, simply + return DECL. */ + +static tree +get_class_access_diagnostic_decl (tree parent_binfo, tree decl) +{ + /* When a class is denied access to a decl in a baseclass, most of the + time it is because the decl itself was declared as private at the point + of declaration. So, by default, DECL is at fault. + + However, in C++, there are (at least) two situations in which a decl + can be private even though it was not originally defined as such. */ + + /* These two situations only apply if a baseclass had private access to + DECL (this function is only called if that is the case). We must however + also ensure that the reason the parent had private access wasn't simply + because it was declared as private in the parent. */ + if (context_for_name_lookup (decl) == BINFO_TYPE (parent_binfo)) + return decl; + + /* 1. If the "using" keyword is used to inherit DECL within a baseclass, + this may cause DECL to be private, so we return the using statement as + the source of the problem. + + Scan the fields of PARENT_BINFO and see if there are any using decls. If + there are, see if they inherit DECL. If they do, that's where DECL was + truly declared private. */ + for (tree parent_field = TYPE_FIELDS (BINFO_TYPE (parent_binfo)); + parent_field; + parent_field = DECL_CHAIN (parent_field)) + { + if (TREE_CODE (parent_field) == USING_DECL) + { + if (cp_tree_equal (decl, + lookup_member (parent_binfo, + DECL_NAME (parent_field), + /*protect=*/0, + /*want_type=*/false, + tf_warning_or_error))) + return parent_field; + } + } + + /* 2. If decl was privately inherited by a baseclass of the current class, + then decl will be inaccessible, even though it may originally have + been accessible to deriving classes. In that case, the fault lies with + the baseclass that used a private inheritance, so we return the + baseclass type as the source of the problem. + + Since this is the last check, we just assume it's true. */ + return TYPE_NAME (BINFO_TYPE (parent_binfo)); +} + /* If the current scope isn't allowed to access DECL along BASETYPE_PATH, give an error, or if we're parsing a function or class template, defer the access check to be performed at instantiation time. @@ -317,34 +373,33 @@ enforce_access (tree basetype_path, tree decl, tree diag_decl, diag_decl = strip_inheriting_ctors (diag_decl); if (complain & tf_error) { - /* We will usually want to point to the same place as - diag_decl but not always. */ + access_kind access_failure_reason = ak_none; + + /* By default, using the decl as the source of the problem will + usually give correct results. */ tree diag_location = diag_decl; - access_kind parent_access = ak_none; - /* See if any of BASETYPE_PATH's parents had private access - to DECL. If they did, that will tell us why we don't. */ + /* However, if a parent of BASETYPE_PATH had private access to decl, + then it actually might be the case that the source of the problem + is not DECL. */ tree parent_binfo = get_parent_with_private_access (decl, - basetype_path); + basetype_path); - /* If a parent had private access, then the diagnostic - location DECL should be that of the parent class, since it - failed to give suitable access by using a private - inheritance. But if DECL was actually defined in the parent, - it wasn't privately inherited, and so we don't need to do - this, and complain_about_access will figure out what to - do. */ - if (parent_binfo != NULL_TREE - && (context_for_name_lookup (decl) - != BINFO_TYPE (parent_binfo))) + /* So if a parent did had private access, then we need to do special + checks to obtain the best diagnostic location decl. */ + if (parent_binfo != NULL_TREE) { - diag_location = TYPE_NAME (BINFO_TYPE (parent_binfo)); - parent_access = ak_private; + diag_location = get_class_access_diagnostic_decl (parent_binfo, + diag_decl); + + /* We also at this point know that the reason access failed was + because decl was private. */ + access_failure_reason = ak_private; } /* Finally, generate an error message. */ complain_about_access (decl, diag_decl, diag_location, true, - parent_access); + access_failure_reason); } if (afi) afi->record_access_failure (basetype_path, decl, diag_decl); diff --git a/gcc/testsuite/g++.dg/pr19377.C b/gcc/testsuite/g++.dg/pr19377.C new file mode 100644 index 00000000000..356329801ac --- /dev/null +++ b/gcc/testsuite/g++.dg/pr19377.C @@ -0,0 +1,21 @@ +/* { dg-do assemble } */ + +class A +{ + protected: + int i; +}; + +class B:public A +{ + private: + using A::i; // { dg-message "declared" } +}; + +class C:public B +{ + void f() + { + A::i = 0; // { dg-error "private" } + } +}; \ No newline at end of file -- 2.25.1