diff options
author | Michael Stahl <michael.stahl@allotropia.de> | 2022-06-16 11:28:05 +0200 |
---|---|---|
committer | Stephan Bergmann <sbergman@redhat.com> | 2022-06-18 09:03:30 +0200 |
commit | 37ec4442d70339dc8ec5fb8e4ec8984420b6e14d (patch) | |
tree | 4175e83124d5fbcb79208966e29b2156093e5106 /include | |
parent | create getter for ScCellValue::mpFormula (diff) | |
download | core-37ec4442d70339dc8ec5fb8e4ec8984420b6e14d.tar.gz core-37ec4442d70339dc8ec5fb8e4ec8984420b6e14d.zip |
o3tl: ensure that the initializer of enumarray contains enough elements
Currently this silently succeeds. Turns out oox already contains some
too-short initializers, let's guess the missing properties are all
invalid.
One downside of the templated parameter pack approach in the enumarray ctor, as
witnessed in vcl/win/window/salframe.cxx, is that argument types can no longer
be implicitly deduced and thus need to be spelled explicitly now in certain
cases.
There were also three uses of enumarry with V being unsigned short (aka
sal_uInt16) that started to cause narrowing conversion errors now and needed to
be adapted: In editeng/source/uno/unonrule.cxx the obvious fix was to use the
proper type for V. In sw/source/core/unocore/unosett.cxx with its odd mix of
saL_Int16 and USHRT_MAX, lets keep things that way for now (probably awaiting
later clean up) and use casts to avoid the implicit narrowing. And in
sw/source/filter/ww8/wrtw8esh.cxx the ESCHER_Prop_* values, while presumably
conceptionally of type sal_uInt16, are plain #defines (thus of type int), so
rather than changing V to int it looked more consistent to explicitly cast the
ESCHER_Prop_* vlaues to sal_uInt16.
(And in tools/source/fsys/urlobj.cxx the poor loplugin:redundantfcast started to
unhelpfully kick in for (only) the first argument now.)
Change-Id: If06c29e673ec7e565e283c6f447889cf1f777cb7
Co-authored-by: Stephan Bergmann <sbergman@redhat.com>
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/135970
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Diffstat (limited to 'include')
-rw-r--r-- | include/o3tl/enumarray.hxx | 17 |
1 files changed, 16 insertions, 1 deletions
diff --git a/include/o3tl/enumarray.hxx b/include/o3tl/enumarray.hxx index a3c09d56bea0..6abfc9ac53ba 100644 --- a/include/o3tl/enumarray.hxx +++ b/include/o3tl/enumarray.hxx @@ -22,6 +22,7 @@ #include <iterator> #include <type_traits> +#include <utility> #include <cassert> namespace o3tl { @@ -55,6 +56,20 @@ public: static const size_type max_index = static_cast<size_type>(E::LAST); + // If this ctor only had the args parameter pack, it would erroneously get picked as a better + // choice than the (implicit) copy ctor (taking a const lvalue reference) when a copy is made + // from a non-const lvalue enumarray; the easiest way to avoid that is the additional arg + // parameter; and to keep things simple that parameter is always passed by const lvalue + // reference for now even if there could be cases where passing it by rvalue reference might be + // beneficial or even necessary if V is a move-only type: + template<typename... T> constexpr enumarray(V const & arg, T && ...args): + detail_values{arg, std::forward<T>(args)...} + { + static_assert(sizeof... (T) == max_index); + } + + enumarray() {} + const V& operator[](E index) const { assert(index>=static_cast<E>(0) && index<=E::LAST); @@ -78,7 +93,7 @@ public: V* data() { return detail_values; } -//private: +private: V detail_values[max_index + 1]; }; |