bugfix: ngx.re: fixed a split() edge-case when using control characters in the regex.

Signed-off-by: Yichun Zhang (agentzh) <[email protected]>

Author: thibaultcha

lib/ngx/re.lua

@@ -50,7 +50,7 @@ local _M = { version = base.version }
 local function re_split_helper(subj, compiled, compile_once, flags, ctx)
     local rc
     do
-        local pos = math_max(ctx.pos - 1, 0)
+        local pos = math_max(ctx.pos, 0)
 
         rc = C.ngx_http_lua_ffi_exec_regex(compiled, flags, subj, #subj, pos)
     end
@@ -87,11 +87,18 @@ local function re_split_helper(subj, compiled, compile_once, flags, ctx)
         return nil, nil, nil
     end
 
+    if from == to then
+        -- empty match, skip to next char
+        ctx.pos = to + 1
+
+    else
+        ctx.pos = to
+    end
+
     -- convert to Lua string indexes
 
     from = from + 1
     to = to + 1
-    ctx.pos = to + 1
 
     -- retrieve the first sub-match capture if any
 
@@ -153,6 +160,9 @@ function _M.split(subj, regex, opts, ctx, max, res)
     local res_idx = 0
     local last_empty_match
 
+    -- update to split_helper PCRE indexes
+    ctx.pos = sub_idx - 1
+
     -- splitting: with and without a max limiter
 
     if max > 0 then

t/re-split.t

@@ -1181,3 +1181,67 @@ len: 5
 qr/\[TRACE   \d+/
 --- no_error_log
 [error]
+
+
+
+=== TEST 36: split by unit separator 1/2 (GH issue lua-nginx-module #1217)
+--- http_config eval: $::HttpConfig
+--- config
+    location /re {
+        content_by_lua_block {
+            local ngx_re = require "ngx.re"
+
+            local subjs = {
+                "1\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+                "1\x1fT\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+                "1\x1fT\x1fT\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+            }
+
+            for _, subj in ipairs(subjs) do
+                local col_list = ngx_re.split(subj, "\\x1f")
+                ngx.say(#col_list, " ", table.concat(col_list, "|"))
+            end
+        }
+    }
+--- request
+GET /re
+--- response_body
+15 1|T|||||||||||||15
+15 1|T|T||||||||||||15
+15 1|T|T|T|||||||||||15
+--- error_log eval
+qr/\[TRACE   \d+/
+--- no_error_log
+[error]
+
+
+
+=== TEST 37: split by unit separator 2/2 (with ctx.pos)
+--- http_config eval: $::HttpConfig
+--- config
+    location /re {
+        content_by_lua_block {
+            local ngx_re = require "ngx.re"
+
+            local subjs = {
+                "1\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+                "1\x1fT\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+                "1\x1fT\x1fT\x1fT\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f\x1f15",
+            }
+
+            for _, subj in ipairs(subjs) do
+                local col_list = ngx_re.split(subj, "\\x1f", nil, { pos = 6 })
+                ngx.say(#col_list, " ", table.concat(col_list, "|"))
+            end
+        }
+    }
+--- request
+GET /re
+--- response_body
+12 |||||||||||15
+13 ||||||||||||15
+13 |T|||||||||||15
+--- error_log eval
+qr/\[TRACE   \d+/
+--- no_error_log
+[error]