Draft: zero pad discontiguous segments for omega scans

partly resolves #183 (closed)

(This will likely taking care of by a new version of gwpy)

gwcelery/tasks/detchar.py

@@ -18,6 +18,7 @@ import getpass
 import glob
 import io
 import json
+import re
 import socket
 import time
 
@@ -135,8 +136,30 @@ def make_omegascan(ifo, t0, durs):
     cache = create_cache(ifo, long_start, long_end)
     strain_name = app.conf['strain_channel_names'][ifo]
     try:
-        ts = TimeSeries.read(cache, strain_name,
-                             start=long_start, end=long_end).astype('float64')
+        try:
+            ts = TimeSeries.read(
+                cache, strain_name, start=long_start,
+                end=long_end
+            ).astype('float64')
+        except ValueError as e:
+            # FIXME: retry with zero padding in case of minor discontiguous
+            # segments see https://git.ligo.org/emfollow/gwcelery/-/issues/183
+            # Typical example involves number of nines at the end for first
+            # timeseries. E.g.
+            # Cannot append discontiguous TimeSeries
+            #     TimeSeries 1 span: [1388412795.5441284 ... 1388412795.999939)
+            #     TimeSeries 2 span: [1388412796.0 ... 1388412797.0)
+            pattern = (
+                r'.*\[([0-9]+\.[0-9]+) ... ([0-9]+\.9{2,}).*\n'
+                r'.*([0-9]+\.0) ... ([0-9]+\.[0-9]+)'
+            )
+            if re.findall(pattern, e.args[0]):
+                ts = TimeSeries.read(
+                    cache, strain_name, pad=0.0,
+                    start=long_start, end=long_end
+                ).astype('float64')
+            else:
+                raise
         # Do q_transforms for the different durations
         qgrams = [ts.q_transform(
             frange=(10, 4096), gps=t0,

[Deep Chatterjee]

A minimum of 2 nines after decimal is from eyeballing the end of segments on sentry.

[Deep Chatterjee]

also the minimum duration used for the scan is 1s, so the omegascan should still "look" ok