I haven't seen any quantitative assessments of the cost of moving to Lua 5.2 from 5.1, but I thought someone might be curious, so I collected some stats while migrating a code base recently.
29,300 lines of Lua
10,400 lines of those are unit tests
Issues encountered:
30: unpack --> table.unpack
9: loadstring --> load
5: pairs ordering assumption
5: loadstring + setfenv --> load
3: loadfile + setfenv --> loadfile
3: maxn
2: require "debug"
1: gsub invalid "%" in replacement string
1: os.execute
1: setfenv (avoidable)
1: match past end of string
1: package.loaders --> package.searchers
Most of these are well documented, but a couple of notes are warranted:
1. The "match past end of string" issue refers to string.find or string.match behaving differently when the "init" argument is beyond the end of the string. Previously, it would match patterns that would match an empty string. Now it returns nil. [FWIW, I prefer the previous behavior, because (a) it preserves this intuitive relation: string.match(str, pat, init) === string.match(str:sub(init), pat); (b) a general backwards compatibility bias, and (c) the change slightly complicates my code in this case. ]
2. The "pairs ordering assumption" issues refer to cases where the code made assumptions about the ordering of keys returned from pairs(). For this code base, these assumptions were made only in the test cases (fortunately), but I wrote an "opairs" ("ordered pairs") alternative to "pairs" for usage in cases like serialization where the ordering becomes observable to clients.
Some subjective comments: Lua 5.2 is clearly a step forward, and the code has improved where it has changed. We have always avoided module() (and most globals for that matter) so that was not an issue in this case. Abolishing setfenv() and getfenv() makes sandboxing cleaner and easier. I look forward to a future Lua that abandons globals completely in favor of manifest upvalues!
-bhk
