makerst: Print status messages to make the CI output clearer

This also removes an unused method.
This commit is contained in:
Hugo Locurcio 2020-05-27 22:24:15 +02:00
parent 52814d111c
commit 9065b7ed4e
No known key found for this signature in database
GPG Key ID: 39E8F8BE30B0A49C
1 changed files with 6 additions and 66 deletions

View File

@ -287,6 +287,8 @@ def main(): # type: () -> None
) )
args = parser.parse_args() args = parser.parse_args()
print("Checking for errors in the XML class reference...")
file_list = [] # type: List[str] file_list = [] # type: List[str]
for path in args.path: for path in args.path:
@ -345,7 +347,10 @@ def main(): # type: () -> None
state.current_class = class_name state.current_class = class_name
make_rst_class(class_def, state, args.dry_run, args.output) make_rst_class(class_def, state, args.dry_run, args.output)
if state.errored: if not state.errored:
print("No errors found.")
else:
print("Errors were found in the class reference XML. Please check the messages above.")
exit(1) exit(1)
@ -561,71 +566,6 @@ def make_rst_class(class_def, state, dry_run, output_dir): # type: (ClassDef, S
index += 1 index += 1
def make_class_list(class_list, columns): # type: (List[str], int) -> None
# This function is no longer used.
f = open("class_list.rst", "w", encoding="utf-8")
col_max = len(class_list) // columns + 1
print(("col max is ", col_max))
fit_columns = [] # type: List[List[str]]
for _ in range(0, columns):
fit_columns.append([])
indexers = [] # type List[str]
last_initial = ""
for idx, name in enumerate(class_list):
col = idx // col_max
if col >= columns:
col = columns - 1
fit_columns[col].append(name)
idx += 1
if name[:1] != last_initial:
indexers.append(name)
last_initial = name[:1]
row_max = 0
f.write("\n")
for n in range(0, columns):
if len(fit_columns[n]) > row_max:
row_max = len(fit_columns[n])
f.write("| ")
for n in range(0, columns):
f.write(" | |")
f.write("\n")
f.write("+")
for n in range(0, columns):
f.write("--+-------+")
f.write("\n")
for r in range(0, row_max):
s = "+ "
for c in range(0, columns):
if r >= len(fit_columns[c]):
continue
classname = fit_columns[c][r]
initial = classname[0]
if classname in indexers:
s += "**" + initial + "** | "
else:
s += " | "
s += "[" + classname + "](class_" + classname.lower() + ") | "
s += "\n"
f.write(s)
for n in range(0, columns):
f.write("--+-------+")
f.write("\n")
f.close()
def escape_rst(text, until_pos=-1): # type: (str) -> str def escape_rst(text, until_pos=-1): # type: (str) -> str
# Escape \ character, otherwise it ends up as an escape character in rst # Escape \ character, otherwise it ends up as an escape character in rst
pos = 0 pos = 0