Files
S.P.L.U.R.T-Station-13/code/__HELPERS/_lists.dm
SandPoot 54641ce201 send
2023-01-23 20:44:28 -03:00

792 lines
23 KiB
Plaintext

/*
* Holds procs to help with list operations
* Contains groups:
* Misc
* Sorting
*/
/*
* Misc
*/
// Generic listoflist safe add and removal macros:
///If value is a list, wrap it in a list so it can be used with list add/remove operations
#define LIST_VALUE_WRAP_LISTS(value) (islist(value) ? list(value) : value)
///Add an untyped item to a list, taking care to handle list items by wrapping them in a list to remove the footgun
#define UNTYPED_LIST_ADD(list, item) (list += LIST_VALUE_WRAP_LISTS(item))
///Remove an untyped item to a list, taking care to handle list items by wrapping them in a list to remove the footgun
#define UNTYPED_LIST_REMOVE(list, item) (list -= LIST_VALUE_WRAP_LISTS(item))
#define LAZYINITLIST(L) if (!L) { L = list(); }
#define UNSETEMPTY(L) if (L && !length(L)) L = null
///Like LAZYCOPY - copies an input list if the list has entries, If it doesn't the assigned list is nulled
#define LAZYLISTDUPLICATE(L) (L ? L.Copy() : null )
#define LAZYREMOVE(L, I) if(L) { L -= I; if(!length(L)) { L = null; } }
//ambition start
#define LAZYCUT(L, S, E) if((length(L) >= S) && (E == 0 || length(L) >= (E - 1))) { L.Cut(S, E); if(!length(L)) { L = null; } }
//ambition end
#define LAZYADD(L, I) if(!L) { L = list(); } L += I;
#define LAZYOR(L, I) if(!L) { L = list(); } L |= I;
#define LAZYFIND(L, V) (L ? L.Find(V) : 0)
///returns L[I] if L exists and I is a valid index of L, runtimes if L is not a list
#define LAZYACCESS(L, I) (L ? (isnum(I) ? (I > 0 && I <= length(L) ? L[I] : null) : L[I]) : null)
#define LAZYSET(L, K, V) if(!L) { L = list(); } L[K] = V;
#define LAZYLEN(L) length(L)
///This is used to add onto lazy assoc list when the value you're adding is a /list/. This one has extra safety over lazyaddassoc because the value could be null (and thus cant be used to += objects)
#define LAZYADDASSOCLIST(L, K, V) if(!L) { L = list(); } L[K] += list(V);
//Sets a list to null
#define LAZYNULL(L) L = null
#define LAZYADDASSOC_TG(L, K, V) if(!L) { L = list(); } L[K] += V;
///This is used to add onto lazy assoc list when the value you're adding is a /list/. This one has extra safety over lazyaddassoc because the value could be null (and thus cant be used to += objects)
#define LAZYADDASSOC(L, K, V) if(!L) { L = list(); } L[K] += list(V);
#define LAZYREMOVEASSOC(L, K, V) if(L) { if(L[K]) { L[K] -= V; if(!length(L[K])) L -= K; } if(!length(L)) L = null; }
#define LAZYACCESSASSOC(L, I, K) L ? L[I] ? L[I][K] ? L[I][K] : null : null : null
#define QDEL_LAZYLIST(L) for(var/I in L) qdel(I); L = null;
//These methods don't null the list
#define LAZYCOPY(L) (L ? L.Copy() : list() ) //Use LAZYLISTDUPLICATE instead if you want it to null with no entries
#define LAZYCLEARLIST(L) if(L) L.Cut() // Consider LAZYNULL instead
#define SANITIZE_LIST(L) ( islist(L) ? L : list() )
#define reverseList(L) reverseRange(L.Copy())
/// Performs an insertion on the given lazy list with the given key and value. If the value already exists, a new one will not be made.
#define LAZYORASSOCLIST(lazy_list, key, value) \
LAZYINITLIST(lazy_list); \
LAZYINITLIST(lazy_list[key]); \
lazy_list[key] |= value;
/// Passed into BINARY_INSERT to compare keys
#define COMPARE_KEY __BIN_LIST[__BIN_MID]
/// Passed into BINARY_INSERT to compare values
#define COMPARE_VALUE __BIN_LIST[__BIN_LIST[__BIN_MID]]
/****
* Binary search sorted insert
* INPUT: Object to be inserted
* LIST: List to insert object into
* TYPECONT: The typepath of the contents of the list
* COMPARE: The object to compare against, usualy the same as INPUT
* COMPARISON: The variable on the objects to compare
* COMPTYPE: How should the values be compared? Either COMPARE_KEY or COMPARE_VALUE.
*/
#define BINARY_INSERT(INPUT, LIST, TYPECONT, COMPARE, COMPARISON, COMPTYPE) \
do {\
var/list/__BIN_LIST = LIST;\
var/__BIN_CTTL = length(__BIN_LIST);\
if(!__BIN_CTTL) {\
__BIN_LIST += INPUT;\
} else {\
var/__BIN_LEFT = 1;\
var/__BIN_RIGHT = __BIN_CTTL;\
var/__BIN_MID = (__BIN_LEFT + __BIN_RIGHT) >> 1;\
var ##TYPECONT/__BIN_ITEM;\
while(__BIN_LEFT < __BIN_RIGHT) {\
__BIN_ITEM = COMPTYPE;\
if(__BIN_ITEM.##COMPARISON <= COMPARE.##COMPARISON) {\
__BIN_LEFT = __BIN_MID + 1;\
} else {\
__BIN_RIGHT = __BIN_MID;\
};\
__BIN_MID = (__BIN_LEFT + __BIN_RIGHT) >> 1;\
};\
__BIN_ITEM = COMPTYPE;\
__BIN_MID = __BIN_ITEM.##COMPARISON > COMPARE.##COMPARISON ? __BIN_MID : __BIN_MID + 1;\
__BIN_LIST.Insert(__BIN_MID, INPUT);\
};\
} while(FALSE)
//Returns a list in plain english as a string
/proc/english_list(list/input, nothing_text = "nothing", and_text = " and ", comma_text = ", ", final_comma_text = "" )
var/total = length(input)
switch(total)
if (0)
return "[nothing_text]"
if (1)
return "[input[1]]"
if (2)
return "[input[1]][and_text][input[2]]"
else
var/output = ""
var/index = 1
while (index < total)
if (index == total - 1)
comma_text = final_comma_text
output += "[input[index]][comma_text]"
index++
return "[output][and_text][input[index]]"
/**
* English_list but associative supporting. Higher overhead.
* @depricated
*/
/proc/english_list_assoc(list/input, nothing_text = "nothing", and_text = " and ", comma_text = ", ", final_comma_text = "")
var/total = length(input)
switch(total)
if (0)
return "[nothing_text]"
if (1)
var/assoc = input[input[1]] == null? "" : " = [input[input[1]]]"
return "[input[1]][assoc]"
if (2)
var/assoc = input[input[1]] == null? "" : " = [input[input[1]]]"
var/assoc2 = input[input[2]] == null? "" : " = [input[input[2]]]"
return "[input[1]][assoc][and_text][input[2]][assoc2]"
else
var/output = ""
var/index = 1
var/assoc
while (index < total)
if (index == total - 1)
comma_text = final_comma_text
assoc = input[input[index]] == null? "" : " = [input[input[index]]]"
output += "[input[index]][assoc][comma_text]"
++index
assoc = input[input[index]] == null? "" : " = [input[input[index]]]"
return "[output][and_text][input[index]]"
//Returns list element or null. Should prevent "index out of bounds" error.
/// @depricated
/proc/listgetindex(list/L, index)
if(LAZYLEN(L))
if(isnum(index) && ISINTEGER(index))
if(ISINRANGE(index,1,L.len))
return L[index]
else if(index in L)
return L[index]
return
//Return either pick(list) or null if list is not of type /list or is empty
/// @depricated
/proc/safepick(list/L)
if(LAZYLEN(L))
return pick(L)
//Checks if the list is empty
/// @depricated
/proc/isemptylist(list/L)
if(!L.len)
return TRUE
return FALSE
//Checks for specific types in a listc
/proc/is_type_in_list(atom/A, list/L)
if(!LAZYLEN(L) || !A)
return FALSE
for(var/type in L)
if(istype(A, type))
return TRUE
return FALSE
//Checks for specific types in specifically structured (Assoc "type" = TRUE) lists ('typecaches')
#define is_type_in_typecache(A, L) (A && length(L) && L[(ispath(A) ? A : A:type)])
//Checks for a string in a list
/proc/is_string_in_list(string, list/L)
if(!LAZYLEN(L) || !string)
return
for(var/V in L)
if(string == V)
return TRUE
return
//Removes a string from a list
/proc/remove_strings_from_list(string, list/L)
if(!LAZYLEN(L) || !string)
return
for(var/V in L)
if(V == string)
L -= V //No return here so that it removes all strings of that type
return
//returns a new list with only atoms that are in typecache L
/proc/typecache_filter_list(list/atoms, list/typecache)
RETURN_TYPE(/list)
. = list()
for(var/atom/A as anything in atoms)
if (typecache[A.type])
. += A
/proc/typecache_filter_list_reverse(list/atoms, list/typecache)
RETURN_TYPE(/list)
. = list()
for(var/atom/atom as anything in atoms)
if(!typecache[atom.type])
. += atom
/proc/typecache_filter_multi_list_exclusion(list/atoms, list/typecache_include, list/typecache_exclude)
. = list()
for(var/atom/atom as anything in atoms)
if(typecache_include[atom.type] && !typecache_exclude[atom.type])
. += atom
///Like typesof() or subtypesof(), but returns a typecache instead of a list
/proc/typecacheof(path, ignore_root_path, only_root_path = FALSE)
if(ispath(path))
var/list/types
var/list/output = list()
if(only_root_path)
output[path] = TRUE
else
types = ignore_root_path ? subtypesof(path) : typesof(path)
for(var/T in types)
output[T] = TRUE
return output
else if(islist(path))
var/list/pathlist = path
var/list/output = list()
if(ignore_root_path)
for(var/current_path in pathlist)
for(var/subtype in subtypesof(current_path))
output[subtype] = TRUE
return output
if(only_root_path)
for(var/current_path in pathlist)
output[current_path] = TRUE
else
for(var/current_path in pathlist)
for(var/subpath in typesof(current_path))
output[subpath] = TRUE
return output
/proc/typecacheof_assoc_list(list/pathlist, ignore_root_path = FALSE)
. = list()
if(!istype(pathlist))
return
for(var/P in pathlist)
var/value = pathlist[P]
for(var/T in (ignore_root_path ? subtypesof(P) : typesof(P)))
.[T] = value
//Empties the list by setting the length to 0. Hopefully the elements get garbage collected
/proc/clearlist(list/list)
if(istype(list))
list.len = 0
return
//Removes any null entries from the list
//Returns TRUE if the list had nulls, FALSE otherwise
/proc/listclearnulls(list/L)
var/start_len = L.len
var/list/N = new(start_len)
L -= N
return L.len < start_len
/*
* Returns list containing all the entries from first list that are not present in second.
* If skiprep = 1, repeated elements are treated as one.
* If either of arguments is not a list, returns null
*/
/proc/difflist(list/first, list/second, skiprep=0)
if(!islist(first) || !islist(second))
return
var/list/result = new
if(skiprep)
for(var/e in first)
if(!(e in result) && !(e in second))
result += e
else
result = first - second
return result
/*
* Returns list containing entries that are in either list but not both.
* If skipref = 1, repeated elements are treated as one.
* If either of arguments is not a list, returns null
*/
/proc/uniquemergelist(list/first, list/second, skiprep=0)
if(!islist(first) || !islist(second))
return
var/list/result = new
if(skiprep)
result = difflist(first, second, skiprep)+difflist(second, first, skiprep)
else
result = first ^ second
return result
//Picks a random element from a list based on a weighting system:
//1. Adds up the total of weights for each element
//2. Gets a number between 1 and that total
//3. For each element in the list, subtracts its weighting from that number
//4. If that makes the number 0 or less, return that element.
//Will output null sometimes if you use decimals (e.g. 0.1 instead of 10) as rand() uses integers, not floats
/proc/pickweight(list/L, base_weight = 1)
var/total = 0
var/item
for (item in L)
if (!L[item])
L[item] = base_weight
total += L[item]
total = rand(base_weight, total)
for (item in L)
total -=L [item]
if (total <= 0)
return item
return null
/proc/pickweightAllowZero(list/L) //The original pickweight proc will sometimes pick entries with zero weight. I'm not sure if changing the original will break anything, so I left it be.
var/total = 0
var/item
for (item in L)
if (!L[item])
L[item] = 0
total += L[item]
total = rand(0, total)
for (item in L)
total -=L [item]
if (total <= 0 && L[item])
return item
return null
//Picks a number of elements from a list based on weight.
//This is highly optimised and good for things like grabbing 200 items from a list of 40,000
//Much more efficient than many pickweight calls
/proc/pickweight_mult(list/L, quantity, base_weight = 1)
//First we total the list as normal
var/total = 0
var/item
for (item in L)
if (!L[item])
L[item] = base_weight
total += L[item]
//Next we will make a list of randomly generated numbers, called Requests
//It is critical that this list be sorted in ascending order, so we will build it in that order
//First one is free, so we start counting at 2
var/list/requests = list(rand(1, total))
for (var/i in 2 to quantity)
//Each time we generate the next request
var/newreq = rand()* total
//We will loop through all existing requests
for (var/j in 1 to requests.len)
//We keep going through the list until we find an element which is bigger than the one we want to add
if (requests[j] > newreq)
//And then we insert the newqreq at that point, pushing everything else forward
requests.Insert(j, newreq)
break
//Now when we get here, we have a list of random numbers sorted in ascending order.
//The length of that list is equal to Quantity passed into this function
//Next we make a list to store results
var/list/results = list()
//Zero the total, we'll reuse it
total = 0
//Now we will iterate forward through the items list, adding each weight to the total
for (item in L)
total += L[item]
//After each item we do a while loop
while (requests.len && total >= requests[1])
//If the total is higher than the value of the first request
results += item //We add this item to the results list
requests.Cut(1,2) //And we cut off the top of the requests list
//This while loop will repeat until the next request is higher than the total.
//The current item might be added to the results list many times, in this process
//By the time we get here:
//Requests will be empty
//Results will have a length of quality
return results
/// Pick a random element from the list and remove it from the list.
/proc/pick_n_take(list/L)
RETURN_TYPE(L[_].type)
if(L.len)
var/picked = rand(1,L.len)
. = L[picked]
L.Cut(picked,picked+1) //Cut is far more efficient that Remove()
//Pick a random element from the list by weight and remove it from the list.
//Result is returned as a list in the format list(key, value)
/proc/pickweight_n_take(list/L, base_weight = 1)
if (L.len)
. = pickweight(L, base_weight)
L.Remove(.)
//Returns the top(last) element from the list and removes it from the list (typical stack function)
/proc/pop(list/L)
if(L.len)
. = L[L.len]
L.len--
/proc/popleft(list/L)
if(L.len)
. = L[1]
L.Cut(1,2)
/proc/sorted_insert(list/L, thing, comparator)
var/pos = L.len
while(pos > 0 && call(comparator)(thing, L[pos]) > 0)
pos--
L.Insert(pos+1, thing)
// Returns the next item in a list
/proc/next_list_item(item, list/L)
var/i
i = L.Find(item)
if(i == L.len)
i = 1
else
i++
return L[i]
// Returns the previous item in a list
/proc/previous_list_item(item, list/L)
var/i
i = L.Find(item)
if(i == 1)
i = L.len
else
i--
return L[i]
//Randomize: Return the list in a random order
/proc/shuffle(list/L)
if(!L)
return
L = L.Copy()
for(var/i=1, i<L.len, ++i)
L.Swap(i,rand(i,L.len))
return L
//same, but returns nothing and acts on list in place
/proc/shuffle_inplace(list/L)
if(!L)
return
for(var/i=1, i<L.len, ++i)
L.Swap(i,rand(i,L.len))
//Return a list with no duplicate entries
/proc/uniqueList(list/L)
. = list()
for(var/i in L)
. |= i
//same, but returns nothing and acts on list in place (also handles associated values properly)
/proc/uniqueList_inplace(list/L)
var/temp = L.Copy()
L.len = 0
for(var/key in temp)
if (isnum(key))
L |= key
else
L[key] = temp[key]
//for sorting clients or mobs by ckey
/proc/sortKey(list/L, order=1)
return sortTim(L, order >= 0 ? /proc/cmp_ckey_asc : /proc/cmp_ckey_dsc)
//Specifically for record datums in a list.
/proc/sortRecord(list/L, field = "name", order = 1)
GLOB.cmp_field = field
return sortTim(L, order >= 0 ? /proc/cmp_records_asc : /proc/cmp_records_dsc)
//any value in a list
/proc/sortList(list/L, cmp=/proc/cmp_text_asc)
return sortTim(L.Copy(), cmp)
//uses sortList() but uses the var's name specifically. This should probably be using mergeAtom() instead
/proc/sortNames(list/L, order=1)
return sortTim(L.Copy(), order >= 0 ? /proc/cmp_name_asc : /proc/cmp_name_dsc)
//Converts a bitfield to a list of numbers (or words if a wordlist is provided)
/proc/bitfield2list(bitfield = 0, list/wordlist)
var/list/r = list()
if(islist(wordlist))
var/max = min(wordlist.len,16)
var/bit = 1
for(var/i=1, i<=max, i++)
if(bitfield & bit)
r += wordlist[i]
bit = bit << 1
else
for(var/bit=1, bit<=65535, bit = bit << 1)
if(bitfield & bit)
r += bit
return r
// Returns the key based on the index
#define KEYBYINDEX(L, index) (((index <= length(L)) && (index > 0)) ? L[index] : null)
/proc/count_by_type(list/L, type)
var/i = 0
for(var/T in L)
if(istype(T, type))
i++
return i
/proc/count_occurences_of_value(list/L, val, limit) //special thanks to salmonsnake
. = 0
for (var/i in 1 to limit)
if (L[i] == val)
.++
/// Returns datum/data/record
/proc/find_record(field, value, list/L)
for(var/datum/data/record/R in L)
if(R.fields[field] == value)
return R
return null
//Move a single element from position fromIndex within a list, to position toIndex
//All elements in the range [1,toIndex) before the move will be before the pivot afterwards
//All elements in the range [toIndex, L.len+1) before the move will be after the pivot afterwards
//In other words, it's as if the range [fromIndex,toIndex) have been rotated using a <<< operation common to other languages.
//fromIndex and toIndex must be in the range [1,L.len+1]
//This will preserve associations ~Carnie
/proc/moveElement(list/L, fromIndex, toIndex)
if(fromIndex == toIndex || fromIndex+1 == toIndex) //no need to move
return
if(fromIndex > toIndex)
++fromIndex //since a null will be inserted before fromIndex, the index needs to be nudged right by one
L.Insert(toIndex, null)
L.Swap(fromIndex, toIndex)
L.Cut(fromIndex, fromIndex+1)
//Move elements [fromIndex,fromIndex+len) to [toIndex-len, toIndex)
//Same as moveElement but for ranges of elements
//This will preserve associations ~Carnie
/proc/moveRange(list/L, fromIndex, toIndex, len=1)
var/distance = abs(toIndex - fromIndex)
if(len >= distance) //there are more elements to be moved than the distance to be moved. Therefore the same result can be achieved (with fewer operations) by moving elements between where we are and where we are going. The result being, our range we are moving is shifted left or right by dist elements
if(fromIndex <= toIndex)
return //no need to move
fromIndex += len //we want to shift left instead of right
for(var/i=0, i<distance, ++i)
L.Insert(fromIndex, null)
L.Swap(fromIndex, toIndex)
L.Cut(toIndex, toIndex+1)
else
if(fromIndex > toIndex)
fromIndex += len
for(var/i=0, i<len, ++i)
L.Insert(toIndex, null)
L.Swap(fromIndex, toIndex)
L.Cut(fromIndex, fromIndex+1)
//Move elements from [fromIndex, fromIndex+len) to [toIndex, toIndex+len)
//Move any elements being overwritten by the move to the now-empty elements, preserving order
//Note: if the two ranges overlap, only the destination order will be preserved fully, since some elements will be within both ranges ~Carnie
/proc/swapRange(list/L, fromIndex, toIndex, len=1)
var/distance = abs(toIndex - fromIndex)
if(len > distance) //there is an overlap, therefore swapping each element will require more swaps than inserting new elements
if(fromIndex < toIndex)
toIndex += len
else
fromIndex += len
for(var/i=0, i<distance, ++i)
L.Insert(fromIndex, null)
L.Swap(fromIndex, toIndex)
L.Cut(toIndex, toIndex+1)
else
if(toIndex > fromIndex)
var/a = toIndex
toIndex = fromIndex
fromIndex = a
for(var/i=0, i<len, ++i)
L.Swap(fromIndex++, toIndex++)
//replaces reverseList ~Carnie
/proc/reverseRange(list/L, start=1, end=0)
if(L.len)
start = start % L.len
end = end % (L.len+1)
if(start <= 0)
start += L.len
if(end <= 0)
end += L.len + 1
--end
while(start < end)
L.Swap(start++,end--)
return L
//return first thing in L which has var/varname == value
//this is typecaste as list/L, but you could actually feed it an atom instead.
//completely safe to use
/proc/getElementByVar(list/L, varname, value)
varname = "[varname]"
for(var/datum/D in L)
if(D.vars.Find(varname))
if(D.vars[varname] == value)
return D
//remove all nulls from a list
/proc/removeNullsFromList(list/L)
while(L.Remove(null))
continue
return L
//Copies a list, and all lists inside it recusively
//Does not copy any other reference type
/proc/deepCopyList(list/l)
if(!islist(l))
return l
. = l.Copy()
for(var/i = 1 to l.len)
var/key = .[i]
if(isnum(key))
// numbers cannot ever be associative keys
continue
var/value = .[key]
if(islist(value))
value = deepCopyList(value)
.[key] = value
if(islist(key))
key = deepCopyList(key)
.[i] = key
.[key] = value
//takes an input_key, as text, and the list of keys already used, outputting a replacement key in the format of "[input_key] ([number_of_duplicates])" if it finds a duplicate
//use this for lists of things that might have the same name, like mobs or objects, that you plan on giving to a player as input
/proc/avoid_assoc_duplicate_keys(input_key, list/used_key_list)
if(!input_key || !istype(used_key_list))
return
if(used_key_list[input_key])
used_key_list[input_key]++
input_key = "[input_key] ([used_key_list[input_key]])"
else
used_key_list[input_key] = 1
return input_key
//Flattens a keyed list into a list of it's contents
/proc/flatten_list(list/key_list)
if(!islist(key_list))
return null
. = list()
for(var/key in key_list)
. |= key_list[key]
/proc/make_associative(list/flat_list)
. = list()
for(var/thing in flat_list)
.[thing] = TRUE
/proc/deep_list2params(list/deep_list)
var/list/L = list()
for(var/i in deep_list)
var/key = i
if(isnum(key))
L += "[key]"
continue
if(islist(key))
key = deep_list2params(key)
else if(!istext(key))
key = "[REF(key)]"
L += "[key]"
var/value = deep_list[key]
if(!isnull(value))
if(islist(value))
value = deep_list2params(value)
else if(!(istext(key) || isnum(key)))
value = "[REF(value)]"
L["[key]"] = "[value]"
return list2params(L)
#define NUMLIST2TEXTLIST(list) splittext(list2params(list), "&")
//Picks from the list, with some safeties, and returns the "default" arg if it fails
#define DEFAULTPICK(L, default) ((islist(L) && length(L)) ? pick(L) : default)
/* Definining a counter as a series of key -> numeric value entries
* All these procs modify in place.
*/
/proc/counterlist_scale(list/L, scalar)
var/list/out = list()
for(var/key in L)
out[key] = L[key] * scalar
. = out
/proc/counterlist_sum(list/L)
. = 0
for(var/key in L)
. += L[key]
/proc/counterlist_normalise(list/L)
var/avg = counterlist_sum(L)
if(avg != 0)
. = counterlist_scale(L, 1 / avg)
else
. = L
/proc/counterlist_combine(list/L1, list/L2)
for(var/key in L2)
var/other_value = L2[key]
if(key in L1)
L1[key] += other_value
else
L1[key] = other_value
/proc/assoc_list_strip_value(list/input)
var/list/ret = list()
for(var/key in input)
ret += key
return ret
/proc/compare_list(list/l,list/d)
if(!islist(l) || !islist(d))
return FALSE
if(l.len != d.len)
return FALSE
for(var/i in 1 to l.len)
if(l[i] != d[i])
return FALSE
return TRUE
#define LAZY_LISTS_OR(left_list, right_list)\
( length(left_list)\
? length(right_list)\
? (left_list | right_list)\
: left_list.Copy()\
: length(right_list)\
? right_list.Copy()\
: null\
)
/proc/is_type_in_ref_list(path, list/L)
if(!ispath(path))//not a path
return
for(var/i in L)
var/datum/D = i
if(!istype(D))//not an usable reference
continue
if(istype(D, path))
return TRUE
/proc/safe_json_encode(list/L, default = "")
. = default
return json_encode(L)
/proc/safe_json_decode(string, default = list())
. = default
return json_decode(string)