vector_algorithms.cpp: introduce namespaces by AlexGuteniev · Pull Request #5450 · microsoft/STL (original) (raw)

Resolves #5418

Why namespaces

The goals were:

• Being able to locate functions using IDE navigation
• Group related stuff to make it clearer what uses what

Both goals are achieved with partial success:

• IDE navigation is way better, but still clobbered with now-empty unnamed namespaces. DevCom-10889095 is about that.
• Grouping works, but some algorithms still use _Find_trats_ or find itself, now it happens across namespaces

Changes

• Move functions so that the related are grouped and unrelated are ungrouped; each group contains its trait structs, implementation functions, and export functions
• Separate traits from _Find_traits_ where makes sense
• Rename existing namespaces and functions to match usual _Naming_style.
• Add new namespaces
• Rename some identifiers to shorter, as they now, in particular _Predicate
• Move comments, change comment, introduce variables to resist silly wrapping
• Remove some using namespace to make it more clear what is used from _Bitmap_details
• Exclude some traits from _ARM64_EC and replace by void or completely remove

Changes are incremental to enable per-commit review.
(It may be complex to review the whole change, as something is both moved and modified)

Algorithms and namespaces

Namespaces:

• _Bitset_from_string, _Bitset_to_string: contain corresponding bitset member functions implementations. These were fine, just renamed.
• _Sorting: minmax_element, minmax, is_sorted_until. These have obvious commonalities, and even referred by the Standard with this name.
• _Removing: remove, unique, remove_copy, unique_copy: They do the same shuffle and share tables for it.
• _Mismatch: only mismatch itself. Originally was among find-like algorithm due to sharing traits. Decided to move out as it does not resemble find, and re-implmenet similar traits separately, which only need some functions from find traits. The use of traits can be completely avoided for this algorithm, but that's another story.
• _Count: only count itself. As originally implemented, it was seen as find-like algorithm, however later it was re-implemented with a dedicated approach. The traits are its own, but still inherit from _Find_traits_, as all functions from them are needed.
• _Find_meow_of: find_first_of and basic_string members that match against a set of characters. Pre-existing namespace, but reorganized a bit, in particular, some using namespace removed.
• _Find_seq: search, find_end, and basic_string members members that match subsequence. These are distinct enough from _Find in that they use SSE4.2 like _Find_meow_of. They still use _Find_traits_, but only for fallbacks. Although if there would ever be 32 and 64 bit implementation, it will likely use _Find_traits_ fully.
• _Find: find, find_last, adjacent_find, search_n, and basic_string members members that match one character, including single-chararcter find_meow_not. They have also much of commonality. search_n stands out a bit, but it still uses the same traits, and generally matches a string against a single character, so that is probably fine to keep it here.
• _Reversing: reverse, reverse_copy. The namespace only contains common fallback

The algorithms that don't have traits or common functions don't need namespaces:

• swap_ranges
• replace

Though I'm sure it is now better with more structure, I'm not sure in particular namespaces usage.