mirror of
https://gitlab2.federez.net/re2o/re2o
synced 2024-11-25 22:22:26 +00:00
Handle queries with "+" operators
This commit is contained in:
parent
fe11867ba9
commit
5eac7f6614
1 changed files with 98 additions and 45 deletions
143
search/views.py
143
search/views.py
|
@ -62,11 +62,22 @@ def is_int(variable):
|
|||
return True
|
||||
|
||||
|
||||
def filter_fields():
|
||||
"""Return the list of fields the search applies to"""
|
||||
return ["users", "clubs", "machines", "factures", "bans", "whitelists", "rooms", "ports", "switches"]
|
||||
|
||||
|
||||
def empty_filters():
|
||||
"""Build empty filters used by Django"""
|
||||
filters = [Q() for f in filter_fields()]
|
||||
|
||||
|
||||
def finish_results(request, results, col, order):
|
||||
"""Sort the results by applying filters and then limit them to the
|
||||
number of max results. Finally add the info of the nmax number of results
|
||||
to the dict"""
|
||||
|
||||
results["users"] += results["clubs"]
|
||||
results["users"] = SortTable.sort(
|
||||
results["users"], col, order, SortTable.USERS_INDEX
|
||||
)
|
||||
|
@ -121,12 +132,6 @@ def search_single_word(word, filters, user, start, end, user_state, aff):
|
|||
)
|
||||
filter_users = (filter_clubs | Q(name__icontains=word))
|
||||
|
||||
if len(word.split(" ")) >= 2:
|
||||
# Assume the room is in 1 word, and the building may be in multiple words
|
||||
building = " ".join(word.split(" ")[:-1])
|
||||
room = word.split(" ")[-1]
|
||||
filter_users |= (Q(room__name__icontains=room) & Q(room__building__name__icontains=building))
|
||||
|
||||
if not User.can_view_all(user)[0]:
|
||||
filter_clubs &= Q(id=user.id)
|
||||
filter_users &= Q(id=user.id)
|
||||
|
@ -211,12 +216,6 @@ def search_single_word(word, filters, user, start, end, user_state, aff):
|
|||
Q(details__icontains=word) | Q(name__icontains=word) | Q(port__details=word)
|
||||
)
|
||||
|
||||
if len(word.split(" ")) >= 2:
|
||||
# Assume the room is in 1 word, and the building may be in multiple words
|
||||
building = " ".join(word.split(" ")[:-1])
|
||||
room = word.split(" ")[-1]
|
||||
filter_rooms |= (Q(name__icontains=room) & Q(building__name__icontains=building))
|
||||
|
||||
filters["rooms"] |= filter_rooms
|
||||
|
||||
# Switch ports
|
||||
|
@ -252,7 +251,7 @@ def search_single_word(word, filters, user, start, end, user_state, aff):
|
|||
|
||||
|
||||
def apply_filters(filters, user, aff):
|
||||
""" Apply the filters constructed by search_single_word.
|
||||
""" Apply the filters constructed by search_single_query.
|
||||
It also takes into account the visual filters defined during
|
||||
the search query.
|
||||
"""
|
||||
|
@ -305,53 +304,117 @@ def apply_filters(filters, user, aff):
|
|||
return results
|
||||
|
||||
|
||||
def get_words(query):
|
||||
"""Function used to split the uery in different words to look for.
|
||||
The rules are simple :
|
||||
def search_single_query(query, filters, user, start, end, user_state, aff):
|
||||
""" Handle different queries an construct the correct filters using
|
||||
search_single_word"""
|
||||
if query["operator"] == "+":
|
||||
# Special queries with "+" operators should use & rather than |
|
||||
for q in query["subqueries"]:
|
||||
# Construct an independent filter for each subquery
|
||||
subfilters = empty_filters()
|
||||
subfilters = search_single_word(q, subfilters, user, start, end, user_state, aff)
|
||||
|
||||
# Apply the new filter
|
||||
for field in filter_fields():
|
||||
filters[field] &= subfilters[field]
|
||||
|
||||
return filters
|
||||
|
||||
# Handle standard queries
|
||||
q = query["text"]
|
||||
return search_single_word(q, filters, user, start, end, user_state, aff)
|
||||
|
||||
|
||||
def create_queries(query):
|
||||
"""Function used to split the query in different words to look for.
|
||||
The rules are the following :
|
||||
- anti-slash ('\\') is used to escape characters
|
||||
- anything between quotation marks ('"') is kept intact (not
|
||||
interpreted as separators) excepts anti-slashes used to escape
|
||||
Values in between quotation marks are not searched accross
|
||||
multiple field in the database (contrary to +)
|
||||
- spaces (' ') and commas (',') are used to separated words
|
||||
- "+" signs are used as "and" operators
|
||||
"""
|
||||
# A dict representing the different queries extracted from the user's text
|
||||
queries = []
|
||||
|
||||
words = []
|
||||
i = 0
|
||||
# Format: {
|
||||
# "text": "", # Content of the query
|
||||
# "operator": None, # Whether a special char ("+") was used
|
||||
# "subqueries": None # When splitting the query in subparts (ex when using "+")
|
||||
# }
|
||||
current_query = None
|
||||
|
||||
# Whether the query is between "
|
||||
keep_intact = False
|
||||
|
||||
# Whether the previous char was a \
|
||||
escaping_char = False
|
||||
|
||||
for char in query:
|
||||
if i >= len(words):
|
||||
if current_query is None:
|
||||
# We are starting a new word
|
||||
words.append("")
|
||||
current_query = { "text": "", "operator": None, "subqueries": None }
|
||||
|
||||
if escaping_char:
|
||||
# The last char war a \ so we escape this char
|
||||
escaping_char = False
|
||||
words[i] += char
|
||||
current_query["text"] += char
|
||||
continue
|
||||
|
||||
if char == "\\":
|
||||
# We need to escape the next char
|
||||
escaping_char = True
|
||||
continue
|
||||
|
||||
if char == '"':
|
||||
# Toogle the keep_intact state, if true, we are between two "
|
||||
keep_intact = not keep_intact
|
||||
continue
|
||||
|
||||
if keep_intact:
|
||||
# If we are between two ", ignore separators
|
||||
words[i] += char
|
||||
current_query["text"] += char
|
||||
continue
|
||||
|
||||
if char == "+":
|
||||
# If we encouter a + outside of ", we replace it with a space
|
||||
words[i] += " "
|
||||
# Can't sart a query with a "+", consider it escaped
|
||||
if len(current_query["text"]) == 0:
|
||||
current_query["text"] = char
|
||||
continue
|
||||
|
||||
# Build a slightly more complicate data structure
|
||||
# This is need for queries like '"A B"+C'
|
||||
if current_query["operator"] is None:
|
||||
current_query["operator"] = "+"
|
||||
current_query["subqueries"] = []
|
||||
|
||||
current_query["subqueries"].append(current_query["text"])
|
||||
current_query["text"] = ""
|
||||
continue
|
||||
|
||||
if char == " " or char == ",":
|
||||
# If we encouter a separator outside of ", we create a new word
|
||||
if words[i] is not "":
|
||||
i += 1
|
||||
continue
|
||||
# If we haven't encountered any special case, add the char to the word
|
||||
words[i] += char
|
||||
|
||||
return words
|
||||
if len(current_query["text"]) == 0:
|
||||
# Discard empty queries
|
||||
continue
|
||||
|
||||
if current_query["operator"] is not None:
|
||||
# If we were building a special structure, finish building it
|
||||
current_query["subqueries"].append(current_query["text"])
|
||||
current_query["text"] = ""
|
||||
|
||||
# Save the query and start a new one
|
||||
queries.append(current_query)
|
||||
current_query = None
|
||||
continue
|
||||
|
||||
# If we haven't encountered any special case, add the char to the word
|
||||
current_query["text"].append(char)
|
||||
|
||||
return queries
|
||||
|
||||
|
||||
def get_results(query, request, params):
|
||||
|
@ -365,22 +428,12 @@ def get_results(query, request, params):
|
|||
user_state = params.get("u", initial_choices(CHOICES_USER))
|
||||
aff = params.get("a", initial_choices(CHOICES_AFF))
|
||||
|
||||
filters = {
|
||||
"users": Q(),
|
||||
"clubs": Q(),
|
||||
"machines": Q(),
|
||||
"factures": Q(),
|
||||
"bans": Q(),
|
||||
"whitelists": Q(),
|
||||
"rooms": Q(),
|
||||
"ports": Q(),
|
||||
"switches": Q(),
|
||||
}
|
||||
filters = empty_filters()
|
||||
|
||||
words = get_words(query)
|
||||
for word in words:
|
||||
filters = search_single_word(
|
||||
word, filters, request.user, start, end, user_state, aff
|
||||
queries = create_queries(query)
|
||||
for query in queries:
|
||||
filters = search_single_query(
|
||||
query, filters, request.user, start, end, user_state, aff
|
||||
)
|
||||
|
||||
results = apply_filters(filters, request.user, aff)
|
||||
|
|
Loading…
Reference in a new issue