Reproduced with FME 2019.2.2.0 (20200103 - Build 19817 - WIN64), both CSV and TEXT readers exhibit the same behavior.

A possible workaround is to use the following code (Python 3.x) in a PythonCreator:

import fmeobjects
import csv

class FeatureCreator(object):
    def __init__(self):
        self.inputfilename = FME_MacroValuesp'INPUT_CSV_FILE']
        self.csvdelimiter = ';' # Modify this as needed
        self.csvquotechar = '"' # Modify this as needed
        self.log = fmeobjects.FMELogFile()
        self.fieldnames =  ]
        
    def input(self, feature):
        pass
        
    def close(self):
        with open(self.inputfilename, 'r') as csvfile:
            csvreader = csv.reader(csvfile, 
                                   delimiter=self.csvdelimiter, 
                                   quotechar=self.csvquotechar)
            for n, row in enumerate(csvreader):
                if n == 0:
                    self.fieldnames = row
                    self.log.logMessageString("Attributes to expose in the PythonCreator:", 
                                              fmeobjects.FME_WARN)
                    for field in row:
                        self.log.logMessageString("    "+field, fmeobjects.FME_WARN)
                else:
                    feature = fmeobjects.FMEFeature()
                    for m, value in enumerate(row):
                        feature.setAttribute(self.fieldnames)m], value)
                    self.pyoutput(feature)

This replaces the CSV reader with the CSV module in the python library, it seems to be able to better handle these filenames. It assumes that the CSV filename is available in the published parameter INPUT_CSV_FILE and that the first line of the CSV contains the attribute names. If necessary, it's fairly easy to modify the script for use in a PythonCaller and to pass the CSV filename as an attribute.

Note that you will have to manually expose the attributes from the CSV file in the PythonCreator. When your run the workspace, it will output some warnings in the FME log (in blue) that shows which attributes to expose.

import fmeobjects
import csv

class FeatureCreator(object):
    def __init__(self):
        self.inputfilename = FME_MacroValuesp'INPUT_CSV_FILE']
        self.csvdelimiter = ';' # Modify this as needed
        self.csvquotechar = '"' # Modify this as needed
        self.log = fmeobjects.FMELogFile()
        self.fieldnames =  ]
        
    def input(self, feature):
        pass
        
    def close(self):
        with open(self.inputfilename, 'r') as csvfile:
            csvreader = csv.reader(csvfile, 
                                   delimiter=self.csvdelimiter, 
                                   quotechar=self.csvquotechar)
            for n, row in enumerate(csvreader):
                if n == 0:
                    self.fieldnames = row
                    self.log.logMessageString("Attributes to expose in the PythonCreator:", 
                                              fmeobjects.FME_WARN)
                    for field in row:
                        self.log.logMessageString("    "+field, fmeobjects.FME_WARN)
                else:
                    feature = fmeobjects.FMEFeature()
                    for m, value in enumerate(row):
                        feature.setAttribute(self.fieldnames)m], value)
                    self.pyoutput(feature)

This is fantastic, thank you @david_r.

Just for extra info, I also had an encoding error in the attribute names so I've just arbitrarily set the attribute names as Column_NUMBER:

feature.setAttribute('Column_' + str(m), value)

and then exposed the attributes of Column_0, Column_1 and Column_2. This makes it a bit more flexible if the column headers in the CSV change names.

Thank you very much for your help, it is greatly appreciated.

feature.setAttribute('Column_' + str(m), value)

Glad to hear the workaround was successful. I've flagged the issue with Safe, hopefully they'll take a closer look.

Thanks @david_r! I can reproduce the behavior mentioned and I've filed (FMEENGINE-63184) in our tracking system. I'll let you and @bi know as soon as the issue has been addressed.