diff --git a/plugins/PostProcessingPlugin/scripts/SearchAndReplace.py b/plugins/PostProcessingPlugin/scripts/SearchAndReplace.py index dabc6b725ab..2c3e1b9ef4b 100644 --- a/plugins/PostProcessingPlugin/scripts/SearchAndReplace.py +++ b/plugins/PostProcessingPlugin/scripts/SearchAndReplace.py @@ -1,16 +1,15 @@ # Copyright (c) 2017 Ghostkeeper # The PostProcessingPlugin is released under the terms of the LGPLv3 or higher. +# Altered by GregValiant (Greg Foresi) February, 2023. +# Added option for a layer search with a Start Layer and an End layer. +# Added 'Ignore StartUp G-code' and 'Ignore Ending G-code' options -import re # To perform the search and replace. - +import re from ..Script import Script - +from UM.Application import Application class SearchAndReplace(Script): - """Performs a search-and-replace on all g-code. - - Due to technical limitations, the search can't cross the border between - layers. + """Performs a search-and-replace on the g-code. """ def getSettingDataString(self): @@ -23,37 +22,175 @@ def getSettingDataString(self): { "search": { - "label": "Search", - "description": "All occurrences of this text will get replaced by the replacement text.", + "label": "Search for:", + "description": "All occurrences of this text (within the search range) will be replaced by the 'Replace with' text. The search string is CASE SPECIFIC so 'LAYER' is not the same as 'layer'.", "type": "str", "default_value": "" }, "replace": { - "label": "Replace", - "description": "The search text will get replaced by this text.", + "label": "Replace with:", + "description": "The 'Search For' text will get replaced by this text. For Multi-Line insertions use the newline character 'backslash plus n' as the delimiter. If your Search term ends with a '\n' remember to add '\n' to the end of the Replace term.", "type": "str", "default_value": "" }, "is_regex": { "label": "Use Regular Expressions", - "description": "When enabled, the search text will be interpreted as a regular expression.", + "description": "When disabled the search string is treated as a simple text string. When enabled, the search text will be re-compiled as a 'regular' python expression.", "type": "bool", "default_value": false + }, + "enable_layer_search": + { + "label": "Enable search within a Layer Range:", + "description": "When enabled, You can choose a Start and End layer for the search. When 'Layer Search' is enabled the StartUp and Ending g-codes are always ignored.", + "type": "bool", + "default_value": false, + "enabled": true + }, + "search_start": + { + "label": "Start S&R at Layer:", + "description": "Use the Cura Preview layer numbering. The Start Layer will be included. Enter '1' to start with gcode ';LAYER:0'. Enter ''-6'' to start with the first layer of a raft.", + "type": "int", + "default_value": 1, + "minimum_value": -6, + "enabled": "enable_layer_search" + }, + "search_end": + { + "label": "Stop S&R at end of Layer:", + "description": "Use the Cura Preview layer numbering. Enter '-1' to search and replace to the end of the file. Enter any other layer number and the replacements will conclude at the end of that layer. If the End Layer is equal to the Start Layer then only that single layer is searched.", + "type": "int", + "default_value": -1, + "minimum_value": -1, + "enabled": "enable_layer_search" + }, + "first_instance_only": + { + "label": "Replace first instance only:", + "description": "When enabled only the first instance is replaced.", + "type": "bool", + "default_value": false, + "enabled": true + }, + "ignore_start": + { + "label": "Ignore StartUp G-code:", + "description": "When enabled the StartUp G-code is unaffected. The StartUp G-code is everything from ';generated with Cura...' to ';LAYER_COUNT:' inclusive.", + "type": "bool", + "default_value": true, + "enabled": "not enable_layer_search" + }, + "ignore_end": + { + "label": "Ignore Ending G-code:", + "description": "When enabled the Ending G-code is unaffected.", + "type": "bool", + "default_value": true, + "enabled": "not enable_layer_search" } } }""" def execute(self, data): + curaApp = Application.getInstance().getGlobalContainerStack() + extruder = curaApp.extruderList + retract_enabled = bool(extruder[0].getProperty("retraction_enable", "value")) + # If retractions are enabled then the CuraEngine inserts a single data item for the retraction at the end of the last layer + # 'top_layer' accounts for that + if retract_enabled: + top_layer = 2 + else: + top_layer = 1 search_string = self.getSettingValueByKey("search") - if not self.getSettingValueByKey("is_regex"): - search_string = re.escape(search_string) #Need to search for the actual string, not as a regex. - search_regex = re.compile(search_string) - replace_string = self.getSettingValueByKey("replace") + is_regex = self.getSettingValueByKey("is_regex") + enable_layer_search = self.getSettingValueByKey("enable_layer_search") + start_layer = self.getSettingValueByKey("search_start") + end_layer = self.getSettingValueByKey("search_end") + ignore_start = self.getSettingValueByKey("ignore_start") + ignore_end = self.getSettingValueByKey("ignore_end") + if enable_layer_search: + ignore_start = True + ignore_end = True + first_instance_only = bool(self.getSettingValueByKey("first_instance_only")) - for layer_number, layer in enumerate(data): - data[layer_number] = re.sub(search_regex, replace_string, layer) #Replace all. + #Find the raft and layer:0 indexes-------------------------------------------------------------------------- + raft_start_index = 0 + layer_0_index = 0 + start_index = 1 + end_index = len(data) + try: + for l_num in range(2,12,1): + layer = data[l_num] + if ";LAYER:-" in layer and raft_start_index == 0: + raft_start_index = l_num + if ";LAYER:0" in layer: + layer_0_index = l_num + break + if raft_start_index == 0: + raft_start_index = layer_0_index + raft_layers = 0 + elif raft_start_index < layer_0_index: + raft_layers = layer_0_index - raft_start_index + else: + raft_layers = 0 + except: + pass + #Determine the actual start and end indexes of the data---------------------------------------------------- + try: + if not enable_layer_search: + if ignore_start: + start_index = 2 + else: + start_index = 1 + if ignore_end: + end_index = len(data) - top_layer + else: + end_index = len(data) + elif enable_layer_search: + if start_layer < 1 and start_layer != -6: + start_index = layer_0_index - raft_layers + elif start_layer == -6: + start_index = 2 + else: + start_index = raft_start_index + start_layer - 1 + if end_layer == -1: + end_index = len(data) - top_layer + else: + end_index = raft_start_index + int(end_layer) + if end_index > len(data) - 1: end_index = len(data) - 1 #For possible user input error + if int(end_index) < int(start_index): end_index = start_index #For possible user input error + except: + start_index = 2 + end_index = len(data) - top_layer - return data \ No newline at end of file + # If "first_instance_only" is enabled: + replaceone = False + if first_instance_only: + if not is_regex: + search_string = re.escape(search_string) + search_regex = re.compile(search_string) + for num in range(start_index, end_index, 1): + layer = data[num] + if re.search(search_regex, layer) and replaceone == False: + data[num] = re.sub(search_regex, replace_string, data[num], 1) + replaceone = True + break + if replaceone: break + return data + + # For all the replacements + if not is_regex: + search_string = re.escape(search_string) + search_regex = re.compile(search_string) + if end_index > start_index: + for index in range(start_index, end_index, 1): + layer = data[index] + data[index] = re.sub(search_regex, replace_string, layer) + elif end_index == start_index: + layer = data[start_index] + data[start_index] = re.sub(search_regex, replace_string, layer) + return data