debian-mirror-gitlab/app/helpers/snippets_helper.rb
2016-09-29 09:46:39 +05:30

85 lines
2.7 KiB
Ruby

module SnippetsHelper
def reliable_snippet_path(snippet, opts = nil)
if snippet.project_id?
namespace_project_snippet_path(snippet.project.namespace,
snippet.project, snippet, opts)
else
snippet_path(snippet, opts)
end
end
# Get an array of line numbers surrounding a matching
# line, bounded by min/max.
#
# @returns Array of line numbers
def bounded_line_numbers(line, min, max, surrounding_lines)
lower = line - surrounding_lines > min ? line - surrounding_lines : min
upper = line + surrounding_lines < max ? line + surrounding_lines : max
(lower..upper).to_a
end
# Returns a sorted set of lines to be included in a snippet preview.
# This ensures matching adjacent lines do not display duplicated
# surrounding code.
#
# @returns Array, unique and sorted.
def matching_lines(lined_content, surrounding_lines, query)
used_lines = []
lined_content.each_with_index do |line, line_number|
used_lines.concat bounded_line_numbers(
line_number,
0,
lined_content.size,
surrounding_lines
) if line.include?(query)
end
used_lines.uniq.sort
end
# 'Chunkify' entire snippet. Splits the snippet data into matching lines +
# surrounding_lines() worth of unmatching lines.
#
# @returns a hash with {snippet_object, snippet_chunks:{data,start_line}}
def chunk_snippet(snippet, query, surrounding_lines = 3)
lined_content = snippet.content.split("\n")
used_lines = matching_lines(lined_content, surrounding_lines, query)
snippet_chunk = []
snippet_chunks = []
snippet_start_line = 0
last_line = -1
# Go through each used line, and add consecutive lines as a single chunk
# to the snippet chunk array.
used_lines.each do |line_number|
if last_line < 0
# Start a new chunk.
snippet_start_line = line_number
snippet_chunk << lined_content[line_number]
elsif last_line == line_number - 1
# Consecutive line, continue chunk.
snippet_chunk << lined_content[line_number]
else
# Non-consecutive line, add chunk to chunk array.
snippet_chunks << {
data: snippet_chunk.join("\n"),
start_line: snippet_start_line + 1
}
# Start a new chunk.
snippet_chunk = [lined_content[line_number]]
snippet_start_line = line_number
end
last_line = line_number
end
# Add final chunk to chunk array
snippet_chunks << {
data: snippet_chunk.join("\n"),
start_line: snippet_start_line + 1
}
# Return snippet with chunk array
{ snippet_object: snippet, snippet_chunks: snippet_chunks }
end
end