[run selector] improve performance when selecting object by time

run_selector.py

@@ -317,7 +317,8 @@ class RunSelector:
 
             def try_load_info(num):
                 try:
-                    self.info[run][num] = self.load_info(run, num)
+                    if num not in self.info[run]:
+                        self.info[run][num] = self.load_info(run, num)
                     success = True
                 except (IOError, AttributeError):
                     success = False
@@ -340,6 +341,51 @@ class RunSelector:
                 time_code = self.info[run][num]["time"]
                 return time_code * self.info[run][num]["unit_time"].express(unit_time)
 
+        # -- A function to search a given time using dichotomy
+        def search(nums, time, position="closest"):
+
+            while len(nums) > 0 and not try_load_info(nums[0]):
+                del nums[0]
+            while len(nums) > 0 and not try_load_info(nums[-1]):
+                del nums[-1]
+
+            if len(nums) == 0:
+                return None
+            ileft, iright = 0, len(nums) - 1
+            if get_time(nums[ileft]) >= time:
+                if position in ["closest", "right"]:
+                    return ileft
+                else:
+                    return None
+            if get_time(nums[iright]) < time:
+                if position in ["closest", "left"]:
+                    return iright
+                else:
+                    return None
+
+            while iright - ileft > 1:
+                imid = (ileft + iright) // 2
+                while not try_load_info(nums[imid]):
+                    del nums[imid]
+                    iright -= 1
+                    imid = (ileft + iright) // 2
+                if get_time(nums[imid]) < time:
+                    ileft = imid
+                else:
+                    iright = imid
+
+            if position == "left":
+                return ileft
+            elif position == "right":
+                return nums[iright]
+            else:
+                dleft = np.abs(get_time(nums[ileft]) - time)
+                dright = np.abs(get_time(nums[iright]) - time)
+                if dleft <= dright:
+                    return ileft
+                else:
+                    return iright
+
         # -- Get the list of seemingly available snapshots on the disk or already selected --
         if do_tests:
             names = glob.glob(
@@ -357,7 +403,7 @@ class RunSelector:
         if isinstance(in_nums, list):
             nums = list(filter(lambda n: n in nums, in_nums))
 
-        nums = np.sort(nums)
+        nums.sort()
 
         # --  Select either the first or last output from the list, or all the valid ones --
         if in_nums == "first":
@@ -376,30 +422,33 @@ class RunSelector:
                 nums = [nums[i]]
             else:
                 nums = []
-        else:
-            nums = list(filter(try_load_info, nums))
 
         # -- Select according to time --
         if time_min is not None:
-            nums = list(filter(lambda n: get_time(n) >= time_min, nums))
+            imin = search(nums, time_min, "right")
+            if imin is not None:
+                nums = nums[imin:]
+            else:
+                nums = []
         if time_max is not None:
-            nums = list(filter(lambda n: get_time(n) <= time_max, nums))
+            imax = search(nums, time_min, "left")
+            if imax is not None:
+                nums = nums[: imax + 1]
 
         if time is not None and len(nums) > 0:
             filtered_nums = []
             if not isinstance(time, list):
                 time = [time]
-            # Get time for all already selected nums
-            time_all = np.asarray([[get_time(n), n] for n in nums])
 
             # For all times provided by the user, select the output closer to it
             for t in time:
-                # Index of this output in the time_all array
-                idx = (np.abs(time_all[:, 0] - t)).argmin()
-                num = int(time_all[idx, 1])
+                iclose = search(nums, t)
+                num = nums[iclose]
                 # Only add each selected output once
                 if num not in filtered_nums:
                     filtered_nums.append(num)
             nums = filtered_nums
+        else:
+            nums = list(filter(try_load_info, nums))
 
         return nums