Uncommnet warnings

This commit is contained in:
Richard Mahn 2022-12-02 17:44:02 -07:00
parent 04388499e6
commit 4e30e112de
1 changed files with 10 additions and 9 deletions

View File

@ -78,8 +78,8 @@ def get_input_fields(input_folderpath:str, BBB:str) -> Tuple[str,str,str,str,str
if line.isdigit():
newC = line
# if int(line) != int(C)+1:
# print(f"\nWARNING at line {line_number}: Chapter number is not increasing as expected: moving from {C} to {newC}")
if int(line) != int(C)+1:
print(f"\nWARNING at line {line_number}: Chapter number is not increasing as expected: moving from {C} to {newC}")
V = '0'
C = newC
glQuote = note = verseText = ''
@ -88,8 +88,8 @@ def get_input_fields(input_folderpath:str, BBB:str) -> Tuple[str,str,str,str,str
if line.startswith(f'{Bbb} {C}:'):
parts = line.split(' ')
newV = parts[1].split(':')[1]
# if int(newV) != int(V)+1:
# print(f"\nWARNING at line {line_number}: Verse number is not increasing as expected: moving from {V} to {newV}")
if int(newV) != int(V)+1:
print(f"\nWARNING at line {line_number}: Verse number is not increasing as expected: moving from {V} to {newV}")
V = newV
verseText = ' '.join(parts[2:])
occurrences = {}
@ -175,8 +175,8 @@ def convert_MSWrd_TN_TSV(input_folderpath:str, output_folderpath:str, BBB:str, n
if (gl_quote.startswith("'")): gl_quote = f'{gl_quote[1:]}'
if (gl_quote.endswith("'")): gl_quote = f'{gl_quote[:-1]}'
gl_quote = gl_quote.replace('" ', '').replace(' "', '').replace("' ", ' ').replace(" '", ' ').replace("'s", 's')
# if '"' in gl_quote or "'" in gl_quote:
# print(f"\nWARNING at {BBB} {C}:{V}: glQuote still has straight quote marks: '{gl_quote}'")
if '"' in gl_quote or "'" in gl_quote:
print(f"\nWARNING at {BBB} {C}:{V}: glQuote still has straight quote marks: '{gl_quote}'")
note = note.strip()
if (note.startswith('"')): note = f'{note[1:]}'
@ -185,8 +185,8 @@ def convert_MSWrd_TN_TSV(input_folderpath:str, output_folderpath:str, BBB:str, n
.replace('".', '”.').replace('",', '”,') \
.replace('("', '(“').replace('")', '”)') \
.replace("' ", ' ').replace(" '", ' ').replace("'s", 's')
# if '"' in note or "'" in note:
# print(f"\nWARNING at {BBB} {C}:{V}: note still has straight quote marks: '{note}'")
if '"' in note or "'" in note:
print(f"\nWARNING at {BBB} {C}:{V}: note still has straight quote marks: '{note}'")
temp_output_TSV_file.write(f'{BBB}\t{C}\t{V}\t{generated_id}\t{support_reference}\t{orig_quote}\t{occurrence}\t{gl_quote}\t{note}\n')
@ -200,7 +200,8 @@ def convert_MSWrd_TN_TSV(input_folderpath:str, output_folderpath:str, BBB:str, n
if completed_process_result.stderr:
print(f" Proskomma {BBB} error output was:\n{completed_process_result.stderr.decode()}")
proskomma_output_string = completed_process_result.stdout.decode()
# print(f"Proskomma {BBB} output was: {proskomma_output_string}") # For debugging JS helper program only
print(f"Proskomma {BBB} output was: {proskomma_output_string}") # For debugging JS helper program only
exit(1)
output_lines = proskomma_output_string.split('\n')
if output_lines:
# Log any errors that occurred -- not really needed now coz they go to stderr