[ntuple] Add GetView<void> with type name string or std::type_info argument
#18185
Conversation
Test Results 19 files 19 suites 4d 14h 35m 3s ⏱️ For more details on these failures, see this check. Results for commit dec2acd. ♻️ This comment has been updated with latest results. |
| template <typename T> | ||
| RNTupleView<T> GetView(ROOT::DescriptorId_t fieldId, std::shared_ptr<T> objPtr, std::string_view typeName) | ||
| { | ||
| static_assert(std::is_void_v<T>, "calling GetView with an type name string is only allowed for [T = void]"); |
There was a problem hiding this comment.
| static_assert(std::is_void_v<T>, "calling GetView with an type name string is only allowed for [T = void]"); | |
| static_assert(std::is_void_v<T>, "calling GetView with a type name string is only allowed for [T = void]"); |
| template <typename T> | ||
| RNTupleView<T> GetView(ROOT::DescriptorId_t fieldId, T *rawPtr, std::string_view typeName) | ||
| { | ||
| static_assert(std::is_void_v<T>, "calling GetView with an type name string is only allowed for [T = void]"); |
There was a problem hiding this comment.
| static_assert(std::is_void_v<T>, "calling GetView with an type name string is only allowed for [T = void]"); | |
| static_assert(std::is_void_v<T>, "calling GetView with a type name string is only allowed for [T = void]"); |
enirolf
force-pushed
the
ntuple-view-typename
branch
from
6f2228b to
bb4d08e
Compare
enirolf
changed the title
GetView<void> with type name string argumentGetView<void> with type name string or std::type_info argument
enirolf
force-pushed
the
ntuple-view-typename
branch
from
bb4d08e to
51503fd
Compare
enirolf
force-pushed
the
ntuple-view-typename
branch
from
51503fd to
7ed3e89
Compare
Enables passing a type name string to `GetView<void>` to signal that the pointer has a different underlying type than the on-disk field's type. The field used by `RNTupleView` will be constructed with this type, in turn benefitting from RNTuples built-in type- and bounds-checking.
...which in turn calls the overload that takes a type name string.
enirolf
force-pushed
the
ntuple-view-typename
branch
from
7ed3e89 to
dec2acd
Compare
| template <typename T> | ||
| ROOT::RNTupleView<T> GetView(ROOT::DescriptorId_t fieldId, std::shared_ptr<T> objPtr, std::string_view typeName) | ||
| { | ||
| static_assert(std::is_void_v<T>, "calling GetView with a type name string is only allowed for [T = void]"); |
There was a problem hiding this comment.
Actually, if we only allow T = void, does it make sense to have this function templated at all? It would deviate from other overloads, but I find it less surprising than the documentation telling me it's a template function and then being presented a static_assert...
There was a problem hiding this comment.
Yeah this is actually something I was asking myself as well when I was writing the docs. Hooking into your other comment, I don't think it is used in framework code so until it is I'm also fine removing it here.
| /// | ||
| /// \sa GetView(std::string_view, std::shared_ptr<T>) | ||
| template <typename T> | ||
| ROOT::RNTupleView<T> GetView(ROOT::DescriptorId_t fieldId, std::shared_ptr<T> objPtr, std::string_view typeName) |
There was a problem hiding this comment.
... and in that light, do we expect usage of std::shared_ptr<void> from framework code?
Enables passing a type name string to
GetView<void>to signal that the pointer has a different underlying type than the on-disk field's type. The field used byRNTupleViewwill be constructed with this type, in turn benefitting from RNTuples built-in type- and bounds-checking.To illustrate further, assuming
myIntis astd::uint64_ton-disk:GetView<void>will currently always reconstruct the field used for loading the values from the on-disk descriptor. However, in the example above this can lead to undefined behavior when the actual values on disk don't fit instd::int32_t. The change proposed in this PR enables passing the type name string toGetView<void>, which is then used for the construction of the field instead, leveraging RNTuple's internal type- and bounds-checking:This feature was requested by and discussed with ATLAS.