Skip to the navigation links
Last modified: 14 November 2023

URL: https://cxc.cfa.harvard.edu/ciao/bugs/dm.html

Bugs: Data Model


Table of Contents

General

Filtering Data

ASCII Kernel

Binning & Rebinning Images


Bugs

General

    Linear transforms have an extra 0.5 bin shift applied.

    When creating a file with a Linear WCS, the CXCDM adjusts the transform parameters to force a half bin offset. While mathematically consistent with the values input, the result may be confusing especially when the transform is well known, eg °C to °F.

    -------------------------------------------------------------------------------
    World Coord Transforms for Columns in Table Block simple2.dat
    --------------------------------------------------------------------------------
    
    ColNo Name
    2: tempF = +32.90 [degree F] +1.80 * (tempC -0.50)
    

    which is mathematically correct, but more commonly written as just

       tempF = +32 + 1.8 * tempC
    

    Incorrect values from CAR transform.

    The WCS library that the DM uses has a problem computing coordinate transforms that involve the CAR transform.

    You may get a seg fault if you try to create a very large image. What constitutes "very large" depends on the data type, but for long and float images, 8192x8192 pixels seems to be the threshold.

    (The image doesn't have to be square; it just needs to have 8192^2 pixels.)

    This condition may be met when the "update=no" option is used. Normally, when you filter a dataset, the data subspace (which describes the boundaries of each column's data and therefore is the intersection of the initial minima and maxima with any subsequent filters) gets updated to reflect the filtering. However, when you give the "update=no" option, you instruct the DM not to update the subspace to reflect the current filter. Therefore, the full ranges for x and y are used in the binning, and you get a 8192x8192 image (and a seg fault, for the reason described above).

    TNULL raw header keywords are not copied to the output file  (19 Aug 2011)

    DM support for FITS TC* keywords

    There are three issues with the generic use of TC* keywords in FITS files read and written by DM tools.

    The first issue: TC*n[A-Z]. The DM function that composes this keyword does not strip off the 'P' from the before adding number and letter to the end.

    Second issue: TCTY* keys are not recognized and therefore are not processed.

    Third issue: DM is not always retaining the T*NAM information when it stores information on the DM descriptor, resulting in output keywords T*TYP instead.

    EXTNAME is forced to be same a HDUNAME on output

    When a FITS file is copied, the EXTNAME is forced on output to be the same as the HDUNAME. Typical Chandra/CIAO files require these to be the same; however, data from other missions and projects may not have the same requirement.

    Updating file in place

    Various CIAO tools including dmappend, dmhedit, dmreadpar, acis_clear_status_bits, and dmgti attempt to modify a files contents in place. In addition users of crates may attempt to write scripts that try to open a file with read+write access.

    Attempting to write to a file that is unwriteable does not generate any waring or error message.

    Files may not be writeable for several reasons. The normal UNIX file permissions may not allow a particular user to modify a file.

    File that are gziped are never writeable, regardless of whether the file itself has file-write permission.

    Standard in and standard out, accessed by the "-" (no quotes) file name are not writeable.

    Individual blocks of data that have filters on the data are not writeable. So

    my_evt.fits[sky=region(ds9.reg)]
    

    is not writeable becuase it is actually filtering the data,

    my_evt.fits[gti3]
    

    is writeable since the filter is applied to select which block in the file.

Filtering Data

    Incorrect results when fitering an image using exclude syntax with full option.

    This bug is triggered when filtering an image with the [exclude ] synatx with a region together with the [opt full] directive to retain the original image size.

    unix% dmcopy "img.fits[exclude sky=region(ciao.reg)][opt full]" filt_img.fits
    

    The pixels outside a box that bounds the region are also filtered out (ie set to 0).

    Workaround:

    In many cases users can easily invert the logic in their region files to avoid needing to use the [exclude ] directive. This can be done any time the region only contains included shapes. For example:

    unix% cat ciao.reg
    circle(...A...)
    circle(...B...)
    circle(...C...)
    ...
    
    unix% cat ciao.reg | awk ' BEGIN {print "field()"} {print "-"$0}' > exclude_ciao.reg
    unix% cat exclude_ciao.reg
    field()
    -circle(...A...)
    -circle(...B...)
    -circle(...C...)
    ...
    

    This file can then be used without needing to use the [exclude ] synatx

    unix% dmcopy "img.fits[sky=region(exclude_ciao.reg)][opt full]" filt_img.fits
    

    This has the advantage of also generally being much faster.

    Region filtering images without specifing axis name(s).

    dm region filtering with unnamed axes may fail for complex regions, eg

    dmlist my_img.fits[circle(10,10,10)-box(0,0,30,30)]
    

    The work around is to explicilty use the axis name, eg

    dmlist my_img.fits[sky=circle(10,10,10)-box(0,0,30,30)]
    

    Filtering on some WCS columns produces incorrect results  (08 Oct 2012)

    When filtering on WCS columns, the range is taken by converting range of the parent columns and using those as the limits of the WCS columns. When the transform is highly non-linear, eg the TAN-P transform used to go from DETX,DETY to THETA,PHI, this can leads to incorrect limits and incorrect filters. Users who want to filter on WCS columns should give explict ranges and not rely on the computed min/maxes.

    bad%  dmcopy "evt.fits[theta=:1]" 
    good% dmcopy "evt.fits[theta=0:1]"
    

    Creating a vector on-the-fly when region filtering

    When region-filtering images, you can create a vector on the fly from any two axes by using a filter like "(#1,#3)=circle(...)". Although the image is filtered correctly with a temporary vector, the region filter isn't recorded in the subspace. Hence, tools that use the filtered file don't know that pixels outside the filter region are invalid. As a result, dmstat reports no nulls in the filtered image (unless you explicitly tell the DM to set pixels outside the filter to null by using "opt null=...").

    Applying a bit-filter expression to an integer column does not work, nor does it cause an error.

    Using incorrect syntax with the rectangle shape does not fail when filtering.

    For example, setting xmax > xmin and/or ymax > ymin. Instead it appears that the Data Model simply swaps the min and max values.

    Filtering an image on logical coordinates causes problems when the short cut of omitting a number (i.e. to indicate the default value) is used.

    The exit status of dmcopy is also incorrectly set to 0 (success):

    unix% dmcopy "image.fits[#1=1:20,#2=:]" delme.fits
    # DMCOPY (CIAO): [ftColRead]: FITS error 308 bad first element number in 
    dataset image.fits Block 1 PRIMARY
    
    unix% echo $status
    0
    

    Workarounds:

    1. Omit the "#2=:" from the filter

    2. Specify a range for both elements: [#1=1:20,#2=1:20]

    Trying to exclude a region filter with update=no will cause the image to be filtered by the region's bounding box.

    For example:

    unix% dmcopy "acis_img.fits[exclude sky=region(src.fits)][opt full,update=no]" filtered.fits 
    

    The regions are correctly excluded; however, the image is also clipped at the bounding box around all the excluded shapes, so the corners of a few chips are removed.

    Workarounds:

    1. Remove update=no. In this case, the Data Model internally inverts all exclude filters to be an inclusive filter, and correctly filters the image.

      Be aware that this process is much slower if the region is large. In that case, it will also add a large region keyword to the file's header, noticeably slowing down any operation on that file.

    2. For ASCII region files, it is also possible to manually invert the filter in the file. The "field()" region syntax is used to include the entire field, then remove the undesired sources. For instance,

      # Region file format: CIAO version 1.0
      circle(1635.5,4113.5,135.11408)
      circle(3975,4233,20)
      circle(2565.5,4129.5,40)
      circle(2129.5,4007.5,40)
      

      would become

      # Region file format: CIAO version 1.0
      field()
      -circle(1635.5,4113.5,135.11408)
      -circle(3975,4233,20)
      -circle(2565.5,4129.5,40)
      -circle(2129.5,4007.5,40)
      

      and the dmcopy filtering command would be

      unix% dmcopy "acis_img.fits[sky=region(src.ascii)][opt full,update=no]" filtered.fits 
      

    The "or" syntax ("||") doesn't work inside a clause  (15 Apr 2011)

    This example command does not work:

    unix% dmlist "evt2.fits[(ccd_id=5||ccd_id=7),pha=2500:3500]" blocks 
    

    Workaround:

    Rewrite to include the filter conditions in each part of the conditional.

    unix% dmlist "evt2.fits[(ccd_id=7,pha=2500:3500)||(ccd_id=5,pha=2500:3500)]" blocks 
    

    Using one column from a vector column in region filter

    Trying to use one column that is part of a vector column in a region expression can lead to a crash

    % dmlist hrcf04482N003_evt2.fits"[(tg_lam,tg_d)=field()]" counts
    4381489
    # 24359: Received error signal SIGSEGV-segmentation violation.
    # 24359: An invalid memory reference was made.
    # 24359: segmentation fault: DMLIST (1) is:
    exit_upon_error->NULL   

    Where tg_d is part of the rd = (tg_r,tg_d) vector column.

    The only work-around is to create a temporary file that dismantels the vector column by removing the other column

    % dmcopy hrcf04482N003_evt2.fits"[cols -tg_r]" tmp_evt
    % dmcopy tmp_evt"[(tg_lam,tg_d)=field()]" counts
    4381489

    Filtering on array columns is undefined.

    Filter on an array column is not supported. The tool will run; however, the results are unpredictable.

ASCII Kernel

    Error when reading SIMPLE text format with 0 rows.

    SIMPLE ASCII files cannot have 0 rows. If there are 0 rows, then the column header definition row is incorrectly read. For example:

    unix% echo "#foo" > bar
    unix% dmlist bar cols
     
    --------------------------------------------------------------------------------
    Columns for Table Block bar
    --------------------------------------------------------------------------------
     
    ColNo  Name                 Unit        Type             Range
       1   foo                               String[4]                  
    
    unix% dmlist bar data
     
    --------------------------------------------------------------------------------
    Data for Table Block bar
    --------------------------------------------------------------------------------
     
    ROW    foo
     
         1 #foo
    
    
    unix% dmlist bar counts
    1
    

    The SIMPLE ASCII file should have 0 rows and we see from the data option that the column definition row has been read in as data.

    Virtual columns are not supported.  (11 Dec 2007)

    In the DM, you can normally do

    unix% dmlist evt.fits"[cols ra,dec]"  data
    

    even though RA and Dec are just coordinate systems defined on the X and Y columns in the file; the DM applies the transform on the fly. This doesn't work yet for ASCII files.

    DTF-FIXED keyword comments may be truncated.

    DTF-FIXED header lines may be up to 1024 characters long. However, if the keyword is longer than the FITS standard, the comment is truncated.

    unix% input.txt output.dtf'[opt kernel=text/dtf-fixed]'
    

    In input.txt:

    TTYPE14 = 'Class' / LV Class Exo: M = missile [B = tactical ballistic  
    missile (except Redstone) apo=80:200]
    R = research rocket O = orbital LV V = RTV Y = Exo weather rocket X =  
    Big test rocket D= Deep space launch
    

    In output.dtf:

    TTYPE14 = "Class " / LV Class Exo: M = missile [B = tactical  
    ballistic missile (except Redstone) apo
    

    Long column descriptions

    ASCII files may contain keyword and column descriptions longer than 80 characters. This can cause random failures in the DM (unable to open columns, wrong data type, etc).

Binning & Rebinning Images

    Rebinning an image with different values for the two axes causes the coordinate information to be lost

    For example:

    unix% dmcopy acis.img"[bin x=::5,y=::6]" acis5x6.img
    

    Using the same value for both axes works correctly:

    unix% dmcopy acis.img"[bin (x,y)=::5]" acis5.img
    

    Running a DM tool on an image where one of the axes has been filtered results in an error.  (17 Jul 2009)

    unix% dmextract
    Input event file (ccd3.sky4.fits[y=3767:][bin sky=annulus(3786,3767,0:380:4)]):
    Enter output file name (rprof.fits):
    # dmextract (CIAO 4.0 Beta 2): WARNING: Input file, "ccd3.sky4.fits[y=3767:]", 
    has no rows in it.
    
    Bus error
    

    Images without a physical coordinate system

    For images without a physical coordinate system the DM will internally create one. This is not written to the output file which may lead to errors if the image size changes due to spatial filtering as the output file then has a different physical WCS compared to the input file.