summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorBrad Payne (Vir Linden) <vir@lindenlab.com>2020-09-11 21:47:37 +0100
committerBrad Payne (Vir Linden) <vir@lindenlab.com>2020-09-11 21:47:37 +0100
commit822a4fb7d2645a4167b987995a190169133cff42 (patch)
tree2e47ab26c208ff824269c9960a16ea9985197c63
parent1eaa76f2f09557e1fc828775249d6eece236572d (diff)
SL-13705 - Additional cell/column properties specified in excel output
-rw-r--r--scripts/code_tools/modified_strings.py32
1 files changed, 27 insertions, 5 deletions
diff --git a/scripts/code_tools/modified_strings.py b/scripts/code_tools/modified_strings.py
index ec1c050a8f..b77115ed46 100644
--- a/scripts/code_tools/modified_strings.py
+++ b/scripts/code_tools/modified_strings.py
@@ -106,7 +106,7 @@ def should_translate(filename, elt, field, val):
if re.match(r"^\s*\d*\s*x\s*\d*\s*$", val):
#print(val, "matches resolution string, will ignore")
return False
- # "value" is a hairball, mostly used to encode non-display info but a few exceptions
+ # "value" attribute is a hairball, mostly used to encode non-display info but a few exceptions
if field == "value":
if elt.text is not None and len(elt.text) > 0:
#print("value has text, ignoring", ET.tostring(elt))
@@ -218,18 +218,40 @@ def make_translation_spreadsheet(mod_tree, base_tree, lang, args):
data.append([val, transl_val, new_val, filename, name, attr])
all_en_strings.add(val)
rows += 1
- if args.verbose and rows>0:
- print(" ",rows,"rows added")
+ save_as_excel(data, lang)
+
+
+def save_as_excel(data, lang):
+
outfile = "SL_Translations_{}.xlsx".format(lang.upper())
- cols = ["EN", "Previous Translation ({})".format(lang.upper()), "ENTER NEW TRANSLATION ({})".format(lang.upper()), "File", "Element", "Field"]
num_translations = len(data)
+ cols = ["EN", "Previous Translation ({})".format(lang.upper()), "ENTER NEW TRANSLATION ({})".format(lang.upper()), "File", "Element", "Field"]
df = pd.DataFrame(data, columns=cols)
- df.to_excel(outfile, index=False)
+
+ writer = pd.ExcelWriter(outfile, engine='xlsxwriter')
+ df.to_excel(writer, index=False, sheet_name = "Sheet1")
+
+ workbook = writer.book
+ worksheet = writer.sheets['Sheet1']
+
+ cell_format = workbook.add_format({'text_wrap': True})
+
+ # Translators primarily care about columns A-C
+ worksheet.set_column('A:C', 100, cell_format)
+ worksheet.set_column('D:D', 50, cell_format, {'hidden': True})
+ worksheet.set_column('E:F', 30, cell_format, {'hidden': True})
+
+ # Lock the column header in place while scrolling
+ worksheet.freeze_panes(1, 0)
+
+ writer.save()
+
if num_translations>0:
print("Wrote", num_translations, "rows to file", outfile)
else:
print("Nothing to translate,", outfile, "is empty")
+
if __name__ == "__main__":